American University Washington, D.C. 20016

© COPYRIGHT

by

Fernanda R. Rosa

2019

ALL RIGHTS RESERVED

To Dona Flor.

GLOBAL INTERNET INTERCONNECTION INFRASTRUCTURE:

MATERIALITY, CONCEALMENT, AND SURVEILLANCE

IN CONTEMPORARY COMMUNICATION

BY

Fernanda R. Rosa

ABSTRACT

This dissertation addresses the inherent social and political impacts engendered by the materiality of the internet and the societal dynamics of infrastructure. With a focus on the information circulation infrastructure of the internet, this is a study of internet governance and design. Specifically, this dissertation makes a first effort in internet governance scholarship to examine governance by internet exchange points (IXPs). It aims to illuminate sociotechnical aspects of IXPs, and reveal the controversies behind them and the social, political, and public values at stake. For that, the research focuses on the significance of IXPs in terms of three pathways: the rights’ vectors of interconnection deployment, the infrastructural interdependencies between the global North and global South, and the search for bodies in the making of surveillance through internet infrastructure.

Guided by science and technology studies (STS) and actor network theory (ANT) as methods to unveil power relations in loco, and decolonial and feminist studies to build on epistemologies of the South, this dissertation uses qualitative methods, such as ethnography of infrastructure and code ethnography, to investigate primary data, along with quantitative methods to analyze secondary data. The methods involve the analysis of 1) the merging of four

IXPs datasets; 2) Border Gateway Protocol (BGP) data of the largest IXP in the global South, ii

IX.br São Paulo; and in the global North, DE-CIX Frankfurt; and 3) ethnographic data collected in Brazil, Germany and Mexico.

The contributions are made along six chapters. In Chapter 1, previous works on IXPs in network engineering are examined, and the dissertation’s theoretical framework, methodology and methods are presented. Chapter 2 shows how Tseltal and Zapoteco communities, in Oaxaca and Chiapas, Mexico, are infrastructuring their wireless networks to reach the larger internet, and how the very moment of network interconnection becomes key to understanding the role of internet infrastructure to challenge and co-produce societal values. Chapter 3 examines the development of the first IXP in Mexico City and reveals how the narrative of potential social benefits from an IXP are not realized in a market of low telecommunications and internet competition. Consequently, it exposes the resultant disparity and lost link between an IXP embedded in commercial values and the need for internet access by indigenous communities, distant, not only territorially, from that IXP.

In Chapters 4 and 5, the global North and the global South are analyzed symmetrically in the IXP ecosystem, in order to make visible both the consistent concealment of the South in the study of internet infrastructure and how illuminating its role leverages our public understanding of internet interconnection politics. Specifically, these chapters rebut the paradigm of “free flow” of information revealing that, instead, the internet is marked by a one-way flow in which giant

IXPs in the North benefit from both their common interests with big content delivery networks and the lack of infrastructural resources in the South.

Finally, Chapter 6 explores how an IXP becomes an instrument of state surveillance by showing the dynamics of the German state to collect data through DE-CIX, exploiting the design affordances of this giant IXP. By unveiling a case that makes explicit the manufacturing of the

iii

“turn to infrastructure to control information” involving public and private actors, the chapter problematizes the extraterritorial effects of a global North country jurisdiction. It also reframes data packets to be conceptually understood as potential information and the virtualization of our bodies that traverse the information circulation infrastructure of the internet attracting the interest of opportunistic actors. The conclusions make recommendations and suggestions for future studies.

This dissertation moves IXPs to a different “mode of existence,” beyond the ones sustained by network engineering scholarship. It touches on key and urgent contemporaneous communication issues, and seek to bring theoretical, empirical and methodological contributions to internet governance policy and scholarship.

iv

ACKNOWLEDGMENTS

There are many, both human and non-human, to whom I am grateful during this Ph.D. journey.

This dissertation would not be conceived as it is without the inspiring discussions focused on media, technology and democracy that professors, students and colleagues engendered at

American University’s School of Communication (SOC) classes and workshops, and the Internet

Governance Lab events. At AU, I am especially grateful to Patricia Aufderheide and Kathryn

Montgomery for their wisdom and mentorship; to Ericka Menchen-Trevino, Filippo Trevisan and Malini Ranganathan for helping me to envision the future; to Derrick Jefferson and the AU library system—incredible fellows, to Jean McGee for knowing everything, and to Teal

Wrocklage for being always attentive to our needs at SOC.

Ideas are just ideas if they are not cultivated. I would not conduct this research without the support and guidance of Laura DeNardis. Beyond her scholarship, her motivating feedback and her respect for my background, choices and passion was the fuel that I needed along the process. In-person and remote meetings, e-mails soon after my voice messages, and letters of recommendations always delivered on time, no matter what, were indications that I was never alone and that I should “keep going,” as she used to say.

I am also grateful to Aram Sinnreich, with whom I worked as research assistant, teaching assistant, shared panels, and from whom I learned a lot. Nanette S. Levinson and Virgilio

Almeida gave me the privilege to work with them and were inspirations during the whole path.

Ethnography research is not only exciting, but it is dependent on numerous resources and supporters. I am grateful for the American University Doctoral Student Research Award in 2017, the Google Policy Fellowship in 2017, and Columbia University’s School of International and

Public Affairs and Carnegie Corporation of New York Research Award in 2016 and 2017. Also, v

during my field research, I had the privilege to get to know many fantastic people infrastructuring the internet. For their invaluable support, I thank especially Alvaro Arroyo,

Francisco Badaró, Hans Ludwing Reyes Chávez, Tiago Gonçalves, Humberto Morales, Felipe de

Jesús Santos Salazar and Acaxao Tecnología’s team, and Julio Sirota and for their time, patience and interest in person as well as through continuous conversations on WhatsApp, which facilitated greatly this research work. In the field, I cannot say thanks enough to Thiago Bomfim and PoP-BA’s team, Carlos Casasús and IX.mx’s team, Hartmut Richard Glaser and NIC.br’s team, Mariano Gómez and Colectivo Ik'taK'op’s team, Osvaldo Martínez, Joaquin Yescas

Martínez, Dennis Jair Mendonza and Antonio Moreiras and IX.br’s team. Thanks also to Klaus

Landefeld and Thomas King from DE-CIX and team for pivotal conversations, and Jon Hjembo, from Telegeography, Arnold Nipper from PeeringDB and Bill Woodcock and PCH’s team for timely support.

For sharing their knowledge and also introducing me to key people, I thank Alexandre

Barbosa, Peter Bloom and Rizhomatica’s team, Loreto Bravo (Maca), Jane Coffin, Néyder

Darín, Vagner Diniz, Luiz Fernando García and R3D’s team, Erick Huerta, Francesca Musiani,

Alejandra Carrillo Olano (Janis) and Hugo Zylberberg. Thanks also to Gato Viejo and all the participants of the worskhop “Charlas con el Diablo” at Rancho Electrónico for the learning and motivation.

For contributing with their expertise, professionalism and generosity, I thank Carolina

Israel, Daniel Ribeiro and Carol, Diego De la Selva, and Patricia Harriman. Diego was responsible for making the German language more accessible to me beyond Google Translate, and Patricia was the first to read this dissertation and make sure it was as clear and accessible as possible, keeping the influences of my Portuguese native language under control.

vi

A special thanks to Zuleika Arashiro, Gustavo Azenha, Marcela Bauer, Juliana Coelho,

Emiliano, Carlos Freire and Gabi, Luísa Abbott Galvão, María Elena González, Marcia Heigasi,

Olga Khrustaleva, Dennis and Leta Kopp, Eliza Kwon, Fabio Lyra, María Luisa Morales, Laura

Moutinho and Paulo, Paula Orlando, Patricia Pavanelli and Marcel, Thais Piffer, Juliana Moraes

Pinheiro, Fernanda Prado, Maria Teresa Rocco, Leonardo Rocha, Caroline Sampaio, Agneris

Sampieri, Ada Siqueira, Glaucia Peres da Silva, Laura Tresca, Mauren Turcatel and Rodrigo, and

Diego Vicentin, Lili and Tomás for their incredible support during this journey.

Friends from the Rede de Pesquisa em Governança da Internet and the Brazilians for

Democracy and Social Justice were true sources of inspiration during the whole time. Thanks also to Mauricio Acuña, Ana Cristina Baldrez and Moacir, Fatima Batista, Ully Busoni, Carlos

Carvalho, Marcia Cavallari, Daniella Diniz, eLNebu, Mauricio Garcia, Malu Giani, María

Alvarez Malvido, Andrea Melloni, Telmila Moura, Vera Ligia Toledo and Gildardo Juárez Vega.

I could not forget how grateful I am to all my public-school teachers and university professors who taught me the most fundamental things for me to become a curious researcher.

And finally, my family. Dayse, Felipe, Jonas, João, Larissa, Luiza and Rodrigo, thanks for your support even if from kilometers of distance. Celia is my cheerleader. Seu Alonso has always the right words, full of proudness. And Dona Flor, well… she is the personification of unconditional love and would always find time to be close, bringing me doce de leite when I most needed it. Obrigada mãe! This dissertation is for you, who since I was a child has taught me how to deal with many facets of inequality with so much dexterity and intelligence.

And to the most important non-human of my life, Bolinha, thank you from my heart.

After two earthquakes together in Mexico, and inordinate changes along these years, including a coup d'etat in Brazil, he has continued here, making things much, much easier.

vii

A Novidade

Gilberto Gil / Bi Ribeiro / Herbert Vianna / João Barone

Álbum: Selvagem, Paralamas do Sucesso, 1986

A novidade veio dar à praia Na qualidade rara de sereia Metade o busto de uma deusa Maia Metade um grande rabo de baleia

A novidade era o máximo Do paradoxo estendido na areia Alguns a desejar seus beijos de deusa Outros a desejar seu rabo pra ceia

Ó mundo tão desigual, Tudo é tão desigual De um lado este carnaval De outro a fome total

E a novidade que seria um sonho O milagre risonho da sereia Virava um pesadelo tão medonho Ali naquela praia, ali na areia

A novidade era a guerra Entre o feliz poeta e o esfomeado Estraçalhando uma sereia bonita Despedaçando o sonho pra cada lado

viii

The Novelty1

Gilberto Gil / Bi Ribeiro / Herbert Vianna / João Barone

Album: Selvagem, Paralamas do Sucesso, 1986

The novelty came to the beach In rare mermaid quality Half the bust of a Mayan goddess Half the tail of a big whale

The novelty was the epitome Of paradox extended on the sand Some desiring her goddess kisses Others wishing for her tail for supper

Oh such an unequal world Everything is so unequal On one side this carnival On the other total hunger

And the novelty that would be a dream The mermaid’s smiling miracle Became such a hideous nightmare There on that beach, there on the sand

The novelty was the war Between the happy poet and the ravenous being Ripping apart a beautiful mermaid Shattering the dream every which way

1 Own translation. I am grateful to Luísa Abbott Galvão and Patricia Harriman for precise translation improvements. ix

TABLE OF CONTENTS

ABSTRACT ...... ii

ACKNOWLEDGMENTS ...... v

LIST OF TABLES ...... xii

LIST OF ILLUSTRATIONS ...... xiii

LIST OF ABBREVIATIONS ...... xv

CHAPTER 1 INTRODUCTION, LITERATURE REVIEW, METHODOLOGY ...... 19

1. Introduction ...... 19 2. The Architectures of IXPs and Underlying Infrastructures ...... 29 3. Theoretical Literature and Conceptual Framework ...... 52 4. Methodology and Methods ...... 76

CHAPTER 2 INFRASTRUCTURING TOWARDS DIGITAL CITIZENSHIP: THE PATHS OF LOCAL INDIGENOUS NETWORKS TO THE GLOBAL INTERNET ...... 99

1. Introduction ...... 99 2. Moving Through a Visible Route from Oaxaca to Santa María Yaviche ...... 101 3. Case 1: Internet Infrastructuring by a Regional Commercial ISP ...... 104 4. Case 2: Internet Infrastructuring by a Municipality ...... 113 5. Case 3: Internet Infrastructuring by a Social Organization ...... 118 6. Security and Privacy ...... 126 7. Ways of Approaching Shared Networks...... 128 8. Values and Power in the Interconnection Arrangements ...... 131 9. Final Considerations ...... 135

CHAPTER 3 THE POLITICS OF AN INTERNET NODE: BUILDING THE FIRST INTERNET EXCHANGE POINT IN MEXICO ...... 137

1. Introduction ...... 137 2. Making an IXP ...... 141 3. The IXP as a Result of Interconnection Politics ...... 150 4. Final Considerations ...... 157

x

CHAPTER 4 THE SOCIETY OF INTERNET EXCHANGE POINTS: ASYMMETRIES BETWEEN THE GLOBAL NORTH AND THE GLOBAL SOUTH ...... 161

1. The Origins of IXPs ...... 162 2. A Study on Internet Exchange Points Datasets...... 169 3. Final Considerations ...... 202

CHAPTER 5 INTERNET INFRASTRUCTURE INTERDEPENDENCIES AND THE CONCEALMENT OF THE GLOBAL SOUTH ...... 203

1. Introduction ...... 203 2. The Free Flow of Information Paradigm ...... 205 3. Methods: Code Ethnography ...... 209 4. Limitations ...... 215 5. Considerations on DE-CIX and IX.br ...... 217 6. Comparing DE-CIX and IX.br BGP Data ...... 224 7. Final Considerations ...... 250

CHAPTER 6 IN SEARCH OF BODIES: SURVEILLANCE AND CONTROL AT INTERNET EXCHANGE POINTS ...... 254

1. Introduction ...... 255 2. Methods...... 259 3. De-Black-Boxing Surveillance through Internet Infrastructure ...... 261 4. Law as a Boundary Object ...... 278 5. The Embodiment of Data Packets ...... 284 6. Final Considerations ...... 289

CHAPTER 7 CONCLUSIONS ...... 292

REFERENCES ...... 298

ANNEX 1...... 334

ANNEX 2...... 335

APPENDIX A ...... 336

APPENDIX B ...... 344

xi

LIST OF TABLES

Table 1. Communities’ Demographic Information ...... 94

Table 2. Duplication Cases in PCH, PeeringDB and Telegeography Datasets ...... 173

Table 3. PCH Dataset per Status and Match Factor ...... 176

Table 4. Comparison of Intersection and Union in Euro-IX, PCH and PeeringDB Datasets Merge ...... 178

Table 5. Comparison of Jaccard and Overlap Indexes in Euro-IX, PCH and PeeringDB Datasets Merge...... 179

Table 6. Synthesis of Euro-IX, PCH, PeeringDB and Telegeography Datasets Merge ...... 181

Table 7. Match Factor per Euro-IX, PCH, PeeringDB and Telegeography Datasets ...... 182

Table 8. IXPs Distribution between the global North and global South ...... 183

Table 9. Top 10 Countries in IXPs Distribution between the global North and global South ... 183

Table 10. Distribution of IXPs with MF 1 IN the global North and global South ...... 184

Table 11. Distribution of IXPs with MF 1 in the global North and global South per Country .. 184

Table 12. Distribution of Countries with Listed IXPs per Continent and per global North and global South ...... 185

Table 13. Distribution of Countries with no IXPs between the global North and global South ...... 189

Table 14. Distribution of Countries with no Listed IXPs per Continent and per global North and global South ...... 189

Table 15. Date Information Source ...... 193

Table 16. DE-CIX Fra Connected Networks per Country from the Global North ...... 226

Table 17. DE-CIX Fra Connected Networks per Country from the Global South ...... 227

Table 18. IX.br Connected Networks per Country from the Global North and Global South ... 229

Table 19. Coverage of Allocated ASNs per Country from the Global North and Global South at IX.br SP and DE-CIX Fra ...... 246

Table 1A. Open Systems Interconnection (OSI) Reference Model and TCP/IP Protocol Stack ...... 334

xii

LIST OF ILLUSTRATIONS

Figure 1. Number of authors per company at the IETF. Source: Reproduction from Arkko, 2019a...... 67

Figure 2. Gross National Income (GNI) by High, Upper Middle, Lower Middle and Low Income Economies. Source: World Bank ...... 98

Figure 3 and Figure 4. The transportation system from Oaxaca to Santa María Yaviche ...... 102

Figures 5 and 6: A household with an external radio antenna and a router inside...... 116

Figure 7, Figure 8, Figure 9 and Figure 10. Bringing internet to the community of San Martín for the first time...... 120

Figure 11. Dates difference in years between PCH and Telegeography ...... 194

Figure 12. IXPs Built per Period – 1993-2017 ...... 195

Figure 13. IXP Built in Africa and South America – 1993-2017 ...... 196

Figure 14. IXP Ecosystem Development – 1993-2017 ...... 198

Figure 15. Map: IXP Ecosystem Development – 1993-2000 ...... 199

Figure 16. Map: IXP Ecosystem Development – 1993-2010 ...... 200

Figure 17. Map: IXP Ecosystem Development – 1993-2017 ...... 201

Figure 18. A Border Gateway Protocol Script. Source: Built on PCH Routing Data ...... 211

Figure 19. Frankfurt Geographic Location. Source: Reproduction from Google Maps...... 217

Figure 20. Fiber Optics Network in Western Europe core in 2001. Source: Reproduction from Rutherford et al. (2004)...... 219

Figure 21. Networks Connected to IX.br SP and DE-CIX Fra. Source: IXPs websites, March 24, 2019...... 224

Figure 22. Networks Connected to DE-CIX Fra and IX.br SP. Source: IXPs websites, March 24, 2019...... 225

Figure 23. Map: IX.br Connected Networks per Region and Brazilian State ...... 232

Figure 24. Map: DE-CIX Connected Networks ...... 233

Figure 25. Map: IX.br Connected Networks per Brazilian States ...... 234

Figure 26. Map: DE-CIX Connected Networks per European Countries ...... 235

xiii

Figure 27. Unique Prefixes and Origin Networks Announced at IX.br SP and DE-CIX Fra. Source: IX.br Looking Glass and PCH Raw Data, March 24, 2019...... 244

Figure 28. Proportion of ASNs Announced at IX.br SP and DE-CIX Fra. Source: IX.br Looking Glass, PCH Raw Data, March 24, 2019, Bates et al. (2019), Maigron (2019)...... 245

Figure 29. Countries Represented in Announcements at IX.br SP and DE-CIX Fra. Source: IX.br Looking Glass and PCH Raw Data, March 24, 2019, Maigron (2019)...... 245

Figure 1A. Changes in Internet Traffic Patterns Representations. Source: Reproduction from Labovitz et al. (2010) p. 78...... 335

xiv

LIST OF ABBREVIATIONS

ABRANET Associação Brasileira de Internet ABRINT Associação Brasileira de Provedores de Internet e Telecomunicações AFIX Africa IXP Association ALTO Application-Layer Traffic Optimization AMS-IX Amsterdam Internet Exchange ANATEL Agência Nacional de Telecomunicações ANID Associação Nacional para Inclusão Digital ANT Actor Network Theory API Application Programming Interface ARPA Advanced Research Projects Agency ARPANET Advanced Research Projects Agency Network AS Autonomous System ASA Army Security Agency ASN Autonomous System Number AS_PATH Autonomous System Path BDSG Bundesdatenschutzgesetz BGP Border Gateway Protocol BLPA Bilateral Peering Arrangement BND Bundesnachrichtendienst BR Brasil BSC Base Station Controller BSI Bundesamt für Sicherheit in der Informationstechnik BTS Base Transceiver Station CABASE Cámara Argentina de Internet CAIDA Center for Applied Internet Data Analysis CALEA Communications Assistance for Law Enforcement Act ccTLD Country Code Top-Level Domain CDN Content Delivery Network CEUXHIDZA Centro Universitario del Pueblo Xhidza CFE Comisión Federal de Electricidad CGNAT Carrier Grade Network Address Translation CGI.br Brazilian Internet Steering Committee CIA Central Intelligence Agency CIO Chief Information Officer CISP Community Internet Service Provider CITI A.C. Consorcio para el Intercambio de Tráfico de Internet CIX Commercial Internet Exchange COFETEL Comisión Federal de Telecomunicaciones CONAPO Consejo Nacional de Población CPE Customer-Premise Equipment CPU Central Process Unit CUDI Corporación Universitaria para el Desarrollo de Internet D3C Dynamic Coalition on Community Connectivity DE-CIX Deutscher Commercial Internet Exchange DHS Department of Homeland Security xv

DIX Digital Equipment Corporation, Intel Corporation and Xerox Corporation DNS Domain Name System EFF Electronic Frontier Foundation FAPESP Fundação de Amparo à Pesquisa do Estado de São Paulo FBI Federal Bureau of Investigation FCC Federal Communications Commission FEPG Federal Engineering Planning Group FIB Forwarding Information Base FIX Federal Internet eXchange FNC Federal Networking Council FRA Frankfurt FTP File Transfer Protocol G77 UNESCO 77 countries Gbps Gigabit per Second GCHQ Government Communications Headquarters GDPR General Data Protection Regulation GNI Gross National Income IANA Internet Assigned Numbers Authority IBGE Brazilian Institute of Geography and Statistics ICANN Internet Corporation for Assigned Names and Numbers ICMP Internet Control Message Protocol ID Identification Field IDB Inter-American Development Bank IEEE Institute of Electrical and Electronics Engineers IETF Internet Engineering Task Force IFT Instituto Federal de Telecomunicaciones IGF Internet Governance Forum IGMP Internet Group Management Protocol INEGI Instituto Nacional de Estadística y Geografia IP Internet Protocol IPv4 Internet Protocol version 4 IPv6 Internet Protocol version 6 ISP Internet Service Provider ITU International Telecommunication Union IX Internet eXchange IXP Internet eXchange Point Kbps Kilobits per Second LACNIC Latin America & Caribbean Network Information Center LAN Local Area Network LGBT Lesbian, Gay, Bisexual, and Transgender LINX London Internet Exchange LIR Local Internet Registry LOCAL_PREF Local Preference MAC Media Access Control Mbps Megabits per Second MED Multi-Exit Discriminator

xvi

MF Match Factor MILNET Military Network MIT Massachusetts Institute of Technology MLPA Multilateral Peering Arrangement MSF Metropolitan Fiber System NAM Non-Alignment Movement NAP Network Access Point NAT Network Address Translation NIC.br Brazilian Network Information Center NIEO New International Economic Order NOC Network Operation Center NSA National Security Agency NSF National Science Foundation NSFNET National Science Foundation Network NWICO New World Information and Communication Order OECD Organisation for Economic Co-operation and Development OP Operating System OSI Open Systems Interconnection PARC Xerox Palo Alto Research Center PCH Packet Clearing House PTT Ponto de Troca de Tráfego de São Paulo QoS Quality of Service R3D Red en Defensa de los Derechos Digitales Redes Redes por la Diversidad, Equidad y Sustentabilidad A.C. RFC Request for Comments RIPE NCC Réseaux IP Européens Network Coordination Centre RIR Regional Internet Registry RNP Rede Nacional de Ensino e Pesquisa SCOT Social Construction of Technology SCT Secretaría de Comunicaciones y Transportes SEDESOL Secretaría de Desarrollo Social SIX Slovak Internet eXchange SLA Service Level Agreement SMS Short Message Service SP Sao Paulo SSL Secure Sockets Layer STS Science and Technology Studies Tbps Terabits per second TCP/IP Transmission Control Protocol/Internet Protocol TFTP Trivial File Transfer Protocol TIC A.C. Telecomunicaciones Indígenas Comunitárias TIWS Telefónica International TLD Top Level Domain TPRC46 46th Research Conference on Communication, Information and Internet Policy TFTP Trivial File Transfer Protocol

xvii

UCLA University of California, Los Angeles U.K. United Kingdom UN United Nations UNESCO United Nations Education, Scientific and Cultural Organization U.S. United States VID Values in Design VLAN Virtual Local Area Network VoIP Voice over Internet Protocol VPN Virtual Private Network VSD Value Sensitive Design WB World Bank WCN Wireless Community Networks WSIS World Summit for the Information Society WWW World Wide Web

xviii

CHAPTER 1

INTRODUCTION, LITERATURE REVIEW, METHODOLOGY

1. Introduction

1.1. Definition of the problem

This study addresses the inherent social and political impacts engendered by the materiality of the internet and the societal dynamics of infrastructure. As a large technical system, the internet is modular, and its modules are locally governed and geographically situated.

Some of this system is built on long-standing infrastructures such as submarine cables that once supported the telegraph and fiber optic cables underlying traditional telecommunication networks. On the other hand, the development of the internet also required the construction of new infrastructures that support its modularity and distributed design. These include tangible technological artifacts over which enormous amounts of data circulate via servers, routers, switches and other components that enable the flow of information online and constitute the internet in a way that is contrary to the common idea of the medium as intangible and transcendent of borders.

The information circulation infrastructure of the internet is the focus of this dissertation.

Every internet design alternative involves dynamics of governance; thus, this is a study of internet governance and design. Specifically, this dissertation focuses on internet nodes that provide the infrastructure for a pivotal network of interconnection arrangements, shedding light on their presences and absences. Throughout internet history, such nodes have been known as

19

Network Access Point (NAP), Internet eXchange (IX), or Internet eXchange Point (IXP). IXPs can be initially understood as physical facilities where several “networks” interconnect to exchange traffic and optimize their costs to transfer data over the internet. These networks are owned by organizations such as telecommunication companies, Internet Service Providers

(ISPs), content providers, and any other businesses that are qualified to be part of IXPs by pursuing a unique identifier—an autonomous system number (ASN)—that allows them to be routed over on the internet. In this regard, any kind of organization that operates a network, including banks, community-based networks and universities, may be part of IXPs, although

Internet Service Providers are their key participants. ISPs collaborate with peers to obtain economic benefits and increase their competitiveness. They establish private agreements with each other to interconnect and that constitute the basis of their routing policies, or how data packets route from one point to another on the global internet. While such interconnection arrangements may physically occur in the companies’ own locations, internet exchange points facilitate these processes and are a crucial vantage point from which to examine internet interconnection politics.

IXPs are unknown artifacts for the majority of internet users. Yet reports of international agencies as the Internet Society, the Organisation for Economic Co-operation and Development, and the World Bank, to name a few, state that they are crucial resources to improve internet quality of service and users’ experience by avoiding unnecessary international traffic, keeping local internet traffic local, and diminishing internet latency and price as a result. Because of that,

IXPs have been associated with tools to support broadband development and decrease the digital divide, without substantial recognition of the limits of such rhetoric on IXPs’ social benefits.

20

On the other hand, in market dynamics hidden from the public, some leading IXPs, based mainly in Europe and the United States, have attracted participants from distant regions of the world and have performed roles quite different from keeping local traffic local. These IXPs have become kinds of giant nodes that attend to both the needs of Internet Service Providers based in underserved regions in terms of affordable infrastructure, interconnection facilities and access to content, as well as the needs of big content providers, usually based in the global North, in constant search of reaching as many ISPs as possible to enable the efficient delivery of their content.

As these leading IXPs attract huge volumes of national and international data, they have become attractive for reasons far beyond the original functions they were designed for, including surveillance and censorship by governments. Although IXPs are a highly privatized layer of the internet, the governance and design of such information circulation infrastructures clearly figure into public interest issues as they support worldwide communication and have been converted into targets for opportunistic actions that affect access to knowledge, freedom of speech and privacy of internet users all over the globe.

Throughout this work, I adopt the word internet as non-capitalized as an indication that as a network of networks, the “Internet” is not a unique entity as the capitalized term implies. From the standpoint of interconnection, there is no different procedure for a network to connect to the

“Internet”; adhering to standards, networks simply connect to each other building an inter-net. In fact, in the early beginning of this ecosystem, the internet was seen as just “‘internetworks’ of basic networks,” whereas today, it is generally seen “as one large (virtual) network” (Dordal,

21

2018, p. 24). Nevertheless, as explained in one of the Network Working Group documents of the

Internet Engineering Task Force (IETF), the protocols used in network communication are the same, regardless of whether the host computers are part of the same network or different and distant networks (Braden, 1989, p. 7), which means that such a virtual and single network is part of the imaginary idea of what the internet is.

In line with that, there is a disseminated editorial standard of capitalizing “Internet” in different media. Reflecting such standard, the word will be capitalized in citations where such form has been used.

1.2. Objectives and Research Questions

This dissertation makes a first effort in internet governance scholarship to examine governance by internet exchange points (IXPs). It aims to illuminate sociotechnical aspects of

IXPs, making visible the controversies behind them and the social, political, and public values at stake. For that, the research brings IXPs into pathways, in Bruno Latour’s terms, establishing three threads to be examined: the rights’ vectors of interconnection deployment, the infrastructural interdependencies between the global North and global South2, and the search for bodies in the making of surveillance through internet infrastructure.

With that, this dissertation moves IXPs to a different “mode of existence,” beyond the ones sustained by network engineering scholarship, touching on key contemporaneous communication issues.

2 I adopt the capitalization of the “South” only, following Boaventura de Sousa Santos, but keep the whole expression capitalized when this is the option of cited sources. 22

Major research questions include what are the implications of IXPs’ governance and design for enacting and constraining communication rights? Following that, this dissertation asks to what extent are indigenous3 communities’ rights and expectations of autonomy affected at the internet interconnection level?

A second area of inquiry is to what extent do interconnection politics look different when examined through global North and global South IXPs vantage points? And following that, what kinds of internet infrastructure interdependencies are visible from IXPs in the global North and the global South when examined symmetrically?

Finally, a third area of inquiry is how can IXPs become technologies of surveillance?

What are the ties that sustain the relations between public and private actors in conducting surveillance through IXPs and what are these actors looking for that makes internet interconnection infrastructure convenient spaces for that?

1.3. Rationale for the Study

The significance of the present research project comes from the fact that IXPs have not yet been examined with a focus on the public, and the guarantee of rights related to the internet, particularly with a global South lens, in which hidden elements of internet infrastructure are brought to light from decolonial and feminist critical perspectives.

Theoretically, this dissertation aims to contribute to understanding the pressing but yet opaque aspects of the “turn to infrastructure to control information” as an increasing

3 I use this term as interviewees and authors refer to themselves and their communities in this way, resignifying its colonial origins and the subaltern meanings that it originally alluded to. 23

phenomenon in internet governance. And as a broader contribution, the present research aims to help decolonize internet infrastructure research, joining voices that imagine a feminist internet, where bodies matter.

One can compare IXPs to airports from where an individual can reach many destinations, and where the research focus has been historically on their engineering capacities and airline companies only, disregarding the individuals impacted, as these are considered distant end-users or airline companies’ customers—not IXPs or airport business! The intriguing point is that similar to airports and airlines, the reason for IXPs and networks to exist are the people, and their needs to transport their virtual bodies from one point to another.

My assumption is that, in the case of IXPs, the bodies circulating are information in written or voice formats encapsulated into data packets—a reasoning derived from the problematizations of cybernetics by Katherine Hayles, and of information by Mark Poster. With that, the topic of study of the present research is framed as the information circulation infrastructure of the internet, shifting the focus of current studies of IXPs where the concept of

“data,” abstract and decontextualized, keeps the internet interconnection infrastructure consistently black-boxed.

This approach aims to leverage public understanding of the internet as a global communication medium, elucidating how it is embedded in values and political narratives from the global North that contribute to obscure consistent power unbalances.

Ultimately, this research seeks to be a robust source of information for the academic and technical communities and policymakers engaged in designing theories and methods, internet functions and policies aligned with social justice values.

24

1.4. Research Pathways and Dissertation Chapters

The first “pathway” that the present analysis on IXPs unveils is focused on the rights’ vectors of interconnection deployment. In Chapter 2, based on infrastructure ethnography methods, this dissertation examines how Tseltal and Zapoteco communities are

“infrastructuring” the internet in their territories. It illuminates, on the one hand, how these actors can be considered codesigners of the internet, a communication infrastructure in continuous deployment, and on the other hand, how the politics of internet interconnection unfavorably impact the rights and autonomy values of these communities, unrepresented voices in the making of privatized internet infrastructure.

It was in Chiapas, Mexico, where Mariano Gómez and his family took me along to install internet access on the rooftop of a house in the small and distant pueblo of San Martin de

Abasolo, bringing the signal from other areas by connecting antennas back to in the back end of the network. There, neighbors followed Mariano’s guidance to transform tree trunks into posts.

These posts would be attached to antennas, fixed up to iron beams, and set up with the support of software installed on a rare laptop in the area to transmit an internet signal through the electromagnetic spectrum. In a country where the internet is considered a human right in the

Constitution, the function to enable the conditions for such rights to be materialized falls on indigenous communities, nature and available infrastructure resources. Along with regional ISPs, and other communities, they are infrastructuring towards a digital citizenship that involves desires and negotiations related to communication infrastructure in ways that expose the limits to reconcile aspirations of communities’ emancipatory values with the constraints of internet interconnection design available for them. 25

Chapter 3 analyzes the formation and maintenance of the first IXP installed in Mexico

City, built under the narratives of national sovereignty and country’s objective of improving connectivity. This analysis sheds light on the networks of meanings related to broadband expansion that emerge around an IXP. This chapter unveils the relations of power that circumscribe the IXP and the realm of internet interconnection, and exposes some of the limits inherently imposed on the expected, and publicly-announced, social benefits from that internet node, unveiling the lost link between the IXP and the communities’ connectivity needs.

Another pathway followed by this dissertation concerns the infrastructural interdependencies between the global North and global South related to internet infrastructure, a major focus of Chapter 4 and 5. In Chapter 4, I analyze the society of IXPs, exploring its precursors’ history, and today’s ecosystem based on data resultant of the merge of the four main datasets available on IXPs. They are Euro-IX, Packet Clearing House, PeeringDB, and

Telegeography. This analysis shows the limitations and differences among current IXP datasets.

It also examines the preponderance of IXPs in the North, and the correlated phenomenon of late deployment of IXPs in the global South, setting the groundwork for asymmetries in internet infrastructure.

Chapter 5 uses the global South lens to redress the concealment of the South in internet infrastructure comparing the biggest IXPs in the world: the first, the Deutscher Commercial

Internet Exchange (DE-CIX) in Frankfurt, Germany (hereafter DE-CIX or DE-CIX Fra), biggest in terms of traffic conveyed through it; and the second, the Ponto de Troca de Tráfego de São

Paulo (PTT-SP or IX.br São Paulo, hereafter IX.br or IX.br SP), in Brazil, biggest in terms of the number of networks connected to it.

26

Through a code ethnography approach focused on Border Gateway Protocol (BGP)4, the analysis inputs are the routes that the networks’ routers share with their peers at IXPs, and through which data packets will reach their destinations on the internet after traversing those nodes. Using the global North and global South categories equally to analyze the nationality of the networks reachable at internet exchange points (IXPs), the research substantiates infrastructure interdependencies, otherwise referred to simply as the internet “free flow” of information. Such interdependencies are only visible when the perspective to “learn from the

South and with the South” as Boaventura de Sousa Santos would say (Santos, 1995, p. 508) is used. Furthermore, the findings that emerge from this code ethnography help explain the geopolitics of today’s internet by adding materiality to it, complementing approaches that alternatively and conversely use geopolitics to explain the internet architecture. These findings build on Internet Governance scholarship, contributing to expand it by including a decolonial lens, and unveiling “knowledge that is ‘otherwise,’” as Sandra Harding would note, in the study of internet governance and design.

The third and final pathway that this dissertation follows is the making of surveillance through internet infrastructure and the search for bodies, the subject of Chapter 6. In dialogue with infrastructure studies and surveillance studies, this chapter seeks to demonstrate “the turn to infrastructure,” as Laura DeNardis conceptualizes it, of actors that exploit the design of IXPs to have access to communication information with political ends. It builds on the concept of

“surveillant assemblages,” as defined by Kevin Haggerty and Richard Ericson, arrangements

4 The Border Gateway Protocol is the language spoken by network routers to communicate with each other and transmit data packets from one point to another on the internet. 27

considered multiple and modular, flexible and unstable. Such assemblages are used following various actors’ desires, which are interpreted here as goals translated into technological artifact functions. The chapter shows how IXPs have inadvertently become part of surveillant assemblages for data collection purpose. The analysis of the surveillance case at DE-CIX

Frankfurt illustrates that in the context of privatized governance—as the internet governance is— state and business actors guided by the desire of control and profit end up inherently joining their resources to execute surveillance practices. The prominence of commercial over public values in the IXP design contributes to that dynamic that feed what Shoshana Zuboff calls “surveillance capitalism,” and that in the present research will be approached looking at neoliberal states. In

Foucault terms, neoliberal states are known for normalizing the use of the market economy to rule political power, and that is what we shall see in the state surveillance practices focused on the internet interconnection infrastructure.

To better understand why an IXP becomes part of a surveillant assemblage and an opportune mechanism to surveillance capitalism, the chapter makes the case that it is necessary to problematize the materiality of data packets that traverse that node. It is only when we understand data packets as “virtual bodies,” a term brought from Katherine Hayles’ scholarship, with intrinsic characteristics connected to real-selves from a non-virtual world, that it becomes clear that so-called massive surveillance is not only highly distributed as the word “massive” implies; it is also focused on specific selves, with certain gender, religion, and nationality characteristics—variables commonly taken as proxy by law enforcement actors to scrutinize their wanted bodies. By illuminating data packet identifiers, communication research advances to distance itself from utopian and opaque views about what the internet encompasses.

28

The dissertation closes with Chapter 7 with recommendations and ideas for further inquiry.

The remainder of the present chapter is organized as follows: section 2 examines the empirical literature on IXP, section 3 provides the theoretical framework in which the dissertation builds on, literature review, section 4 is dedicated to the methodology theories and the methods adopted.

2. The Architectures of IXPs and Underlying Infrastructures

The internet is an arrangement of interconnected private networks, each of them with different types of technical and political control (Abbate, 1999; DeNardis, 2014; Roberts et al.,

2011). Private networks, in this case, comprise a group of computers, cables, routers and other equipment that communicate with each other through common languages, or protocols. A network can be run by an Internet Service Provider (ISP), a content provider, a bank, a community, a governmental agency, a university, and so on, identified as an autonomous system

(or systems) through autonomous system number.5 AT&T, an ISP, and Google, a content provider, are examples of network operators that run autonomous systems. When users send information through their ISP—an e-mail, for example—, it may traverse multiple interconnected networks before arriving to the assigned recipient.

5 For now, an Autonomous system (AS) can be understood as a group of networks that is administered by network operators following a single routing policy to forward packets through the internet (Hawkinson & Bates, 1996). Commonly, a network has numerous autonomous systems under it, but in the present work, unless stated, autonomous systems and networks will be used interchangeably as done in the network engineering scholarship. 29

Thus, although network interconnections are not visible to individuals in general, such operations are ordinary and essential for the internet. They are part of its economy and are commonly administered privately through commercial agreements. This is why internet interconnection has been considered “one of the most privatized areas of the Internet” (DeNardis,

2014, p. 223).

Independent of the kind of agreement in place, the interconnections require physical structures to be established. In this context, while companies’ own locations can be used for that, important interconnection facilities gain relevance: internet exchange points (IXPs). IXPs are generally defined as physical and logical spaces in which numerous autonomous systems can connect to each other to exchange traffic and obtain mutual benefit by amplifying their connectivity and reachability. IXPs are considered network “meeting places” (Chatzis et al.,

2013; Metz, 2001), whose precursors originated in the early days of the internet as will be discussed later.

2.1.IXPs from a Network Engineering Viewpoint

The technological nature of IXPs has inspired academic studies in computer science, business and engineering that use network measurement methods to shed light on the internet topology, the technical optimization of the network, and the economic effects of IXPs. Some studies have also examined public databases to illuminate IXPs regional locations, and the amount of traffic that they attract. It is useful to look at the definitions and findings that these kinds of studies have provided.

30

2.1.1. Definitions of IXPs

Engineers have defined internet exchange points as “a network infrastructure with the purpose to facilitate the exchange of Internet traffic between Autonomous Systems and [that] operate below layer three” (Chatzis et al., 2013, p. 20). Some overlay a more normative view, defining IXPs in regard to the absence of regulation. Rodérick Fanou and colleagues’ work is an example, stating that an IXP is “a shared layer-2 switch fabric environment, with three or more participants, where new participation is not rigorously constrained, and over which the members peer with each other, exchanging customer routes” (Fanou, Valera, et al., 2017, p. 4, emphasis added).6

Another definition of IXP that assumes a normative view, based on regionalization goals, is posited by Danny Perez and colleagues, for whom “An IXP is a shared physical network infrastructure regionally installed with the purpose to facilitate the exchange of Internet traffic between ASes [autonomous systems]” (Perez et al., 2016, p. 3). Explaining the benefits of the regionalization of such infrastructure, they continue: “With the traffic exchange as local as possible between different networks that belong to the same region, the number of hops between

ASes and dependency on transit providers is reduced” (Perez et al., 2016). 7 However, reducing

6 “Layers” are abstractions used by the internet community to conceptually describe a network ecosystem. For that, there are two basic references: the Open Systems Interconnection model, also known as the OSI model, which comprises seven layers, and the TCP/IP Protocol Architecture Model (acronym to Transmission Control Protocol/Internet Protocol), which comprises five layers (Oracle, n.d.-a, n.d.-b). The layer-2 switch that is below layer-3, as mentioned by the authors above, corresponds to the Data Link layer in both models. It provides point-to- point data transfer (Shaw, 2017). For more information on the internet layers, please see Annex 1 and Appendix B part 1. Switches are key devices for receiving and forwarding traffic on the internet and will be further discussed ahead. 7 Hop is a term that refers to the action of a packet passing through a network device such as a router, a switch or another equipment. It indicates how many devices intermediate a request from its source to its destination (Fisher, 31

the number of hops is by no means a straightforward engineering technique, as the commercial arrangements of autonomous systems are the main drivers to define the paths.

Examples of IXP regionalization can be given by Argentina and Brazil, where such a goal has been planned and executed by one core organization responsible for the whole IXP ecosystem in each country: the Cámara Argentina de Internet (CABASE) and the Brazilian

Center for Information and Coordination of dot-BR (NIC.br). Nevertheless, the regionalization model is not a rule for the design of IXPs ecosystems worldwide.8

Such definitions have been developed within a network engineering framework focused on highlighting certain IXP affordances or “imagined affordances,” attributes that “emerge between users’ perceptions, attitudes, and expectations; between the materiality and functionality of technologies; and between the intentions and perceptions of designers” (Nagy & Neff, 2015, p. 5). In the cases above, as presented by these authors, IXPs’ affordances are strictly associated with facilitating node-to-node communication, and in some cases, the exchange of traffic regionally. These approaches are commonly guided by the industry definition that is compiled in the report of the European Association of IXPs where it is stated that: “An Internet Exchange

Point (IXP) is a network facility that enables the interconnection and exchange of Internet traffic between more than two independent Autonomous Systems” (Euro-IX, 2015, p. 3).

2018). It is a kind of measure between source and destination, which should not be confused with distance as one hop can be longer than five hopes, for instance. 8 It is also worth mentioning that there may have been different regionalization designs. According to interviews, unlike Brazil, in Argentina, content hosted in one regional IXP is routed to all the other IXPs across the country through routers to facilitate access to content in regions whose small traffic would not attract content providers. Interestingly, considering the layered-internet framework, whereas IXPs are conceptually located on layer-2 (based on switches), some, as the Argentinian case shows, may work also on layer-3 (based on routers) to overcome economic constraints and to reach access goals. 32

While for social scientists and communication scholars these are opaque descriptions of what IXPs are, they are representative of the common approach of this body of literature on

IXPs. As seen, some definitions express expectations with regard to what an IXP should be, for instance, in terms of the absence of regulatory constraints to participate in IXPs, or yet IXPs’ regional base design. I highlight such expectations because as the following chapters will show, many technical and political aspects complicate what are intended to be general rules.

Before deepening our understanding of definitions from the network engineering scholarship, it is useful to understand the types of commercial agreements that lead to set infrastructure interconnection among autonomous systems.

2.1.2. Peering vs. Transit

An interesting scenario underlies the politics of interconnection, in which increasing competition among Internet Service Providers depends on establishing cooperation to enhance global connectivity (Metz, 2001). That happens primarily through commercial agreements known as “peering,” complemented by customer-provider relationships known as “transit.”

Peering, with its roots in IXPs precursors, is a collaborative relationship that is mutually beneficial for the entities involved due to the sharing of resources. In this relationship, autonomous systems, such as ISPs, allow the peers to have access to their routes and their clients’ routes in search of the same benefits in return. Customarily, this relationship does not have monetary payment involved, because they are based on the assumption of parity between peers, but depending on how imbalanced the traffic is, ISPs can establish paid-peering (Faratin et al., 2008; Metz, 2001). 33

When peering involves two autonomous systems only, it is known as “bilateral peering,” or a Bilateral Peering Arrangement (BLPA). When autonomous systems connect to multiple peers at once, this is known as “multilateral peering,” or a Multilateral Peering Arrangement

(MLPA) (Giotsas et al., 2013; Metz, 2001; Richter et al., 2014). IXPs facilitate any of these arrangements, although a bilateral peering may occur not only at IXPs, but also in private facilities where two players decide to interconnect directly. IXPs are considered public peering facilities, instead. Nevertheless, it is worth noting that being characterized as public is not related with IXPs’ types of administration and governance, which can be public or private, profit or not- for-profit.

Unlike peering, “transit,” is a customer-provider relationship established between two autonomous systems, whereby access to the internet is provided through a financial settlement

(Faratin et al., 2008; Metz, 2001). In this relationship, one wants to buy connectivity while the other one has the infrastructural resources to sell it, but this is a conjunctural situation, which depends on the circumstances: In the 1990s, some ISPs decided to end peering agreements with smaller ISPs and started to peer among themselves only, and to sell transit to the rest (Metz,

2001). They are autonomous to do that.

A common situation for ISPs is to settle peering and transit in a complementary way, as routes of peers and their clients will not give access to the whole internet, requiring transit services (Metz, 2001). In fact, an economic rationale guides autonomous systems’ decisions towards opting for peering or transit. As Chris Metz points out: “ISPs are driven to lower costs, maximize performance, and generate revenue. The choice of where and with whom to peer or transit directly impacts these driving factors” (Metz, 2001, p. 74). On the other hand, inside

34

autonomous systems’ networks there may also occur sibling relationships, when networks under the same administration deploy their own types of private agreements (Dimitropoulos et al.,

2007). Economic drivers are thus the basis of the interconnection dynamics and guide the political decisions that can be seen through the study of IXPs, but most of the time, are overlooked in the current IXP scholarship and policy world.

2.1.3. IXPs as a Vantage Point to Study the Internet

Esteban Carisimo, Hernan Galperin and Jose Ignacio Alvarez-Hamelin classify the works on IXPs according to the measurement methods used to study them. They identify six common approaches that vary in terms of technique and sources: “(i) Using internal vantage points, (ii)

Analyzing IXP's website, (iii) Mining into Public Databases, (iv) ICMP-based measurements9,

(v) BGP-based measurements10 and (vi) Developing platforms using public resources”

(Carisimo et al., 2015).

In this regard, Vasileios Giotsas and colleagues, including Kimberly Claffy (kc claffy), the founder of one of the most important centers for internet measurement in the United States, the Center for Applied Internet Data Analysis (CAIDA), suggest that the sources most currently include traceroutes and BGP data (v and vi options above), and highlight the thought-provoking fact related to the limitations of these sources, in a comprehensive look at the interconnection scenario of peer-to-peer connections. They argue that there is a high level of “incompleteness,”

9 Internet Control Message Protocol (ICMP)-based measurements commonly include datasets based on traceroutes paths. This is possible because ICMP acts as a transport protocol that moves data end to end (Braden, 1989, p. 10). 10 Border Gateway Protocol (BGP)-based measurements use BGP routing tables to examine autonomous systems, their networks and routing paths. For a discussion on this protocol, see Appendix B. 35

not only because of the private policies of internet operators that restrict access to their links to other networks due to commercial purposes, but also because internet exchange points make arrangements among networks difficult to visualize (Giotsas et al., 2013). Network engineering scholars are the main authors of these studies. They show a constant search for creating new mechanisms and explore public sources to circumvent available internet measurements constraints to provide a different views on the internet complexity (e.g. Ager et al., 2012; Giotsas et al., 2013; Lodhi et al., 2014). From outside, the view is that works in this area resemble reports of discovery expeditions, disseminated mainly through specialized conferences.

In terms of the results brought by the internet measurement literature, IXPs are presented as a yet understudied but a useful vantage point for understanding the topology, economy and dynamics of the internet (Chatzis et al., 2013). Actually, studies built on BGP-based measurements after fifteen years of the beginning of the commercial internet have indicated an important shift in the internet topology from a hierarchical to a flat model (Gill et al., 2008;

Labovitz et al., 2010)—a fact that is intrinsically related to the increasing role of IXPs in network connectivity (Dhamdhere & Dovrolis, 2010).11 Authors explain that since the decommission of the National Science Foundation backbone12 and the transition to a commercial internet, a relationship between three kinds of players has historically characterized the internet: Tier-3 networks are Internet Service Providers (ISPs), circumscribed geographically, which provide internet access to end users and originally to content providers. For instance, these can be companies that serve some limited cities or states. Tier-2 networks, on the other hand, are ISPs

11 For a graphical representation of these topologies, see Annex 2. 12 “A backbone network (…) serves a collection of local networks (…) connected to a set of computers such as desktops, servers, or mobile devices (…). The backbone is structured as a mesh so that the loss of any backbone node does not partition the network into isolated, unconnected pieces” (Denning & Martell, 2015, p. 222). 36

with regional scope that connect with each other and that have typically provided internet to the aforementioned Tier-3 ones. Tier-2 companies can be national ISPs, for example, including telco incumbents. Finally, Tier-1 networks constitute a few backbone networks that through commercial agreements have access to the entire internet. For instance, Congent, Level 3,

Verizon and Telefonica International, are part of the small group of well-known Tier-1 companies.

Historically, Tier-2 networks have bought their connection to the larger internet typically from Tier-1. This hierarchical design, built on commercial relationships where Tier-3 networks have been transit customers of Tier-2, which in turn are transit customers of Tier-1 networks, has moved to “a more densely interconnected and less hierarchical inter-domain [inter-network]

Internet topology” (Labovitz et al., 2010). In this new model, instead of the prevalence of transit agreements, or customer-provider relationships, Tier-2 and Tier-3 ISPs as well as content providers establish peering agreements facilitated by internet exchange points’ physical infrastructures, where commercial relationships are established based on collaborative traffic exchange.13

The implications of that are various. In 2008, Philippa Gill and colleagues had already identified a trend in which the traffic of big content providers at that time, Microsoft, Yahoo! and

Google, traversed fewer Tier-1, or Tier-1 hops, than others. Indeed, more than 60% of the biggest content providers did not contain any Tier-1 hops (Gill et al., 2008, p. 6), which indicates that, in such cases, content was being delivered without the use of transit agreements with global networks. Instead, the study found out that major content providers were carrying out their own

13 Network agreements, such as peering and transit, will be better explored ahead. 37

network infrastructure and that the entry points on that were “large centers where carrier hotels or Internet Exchanges exist” (Gill et al., 2008, p. 7). The authors speculate that the motivations for that change could be various, including eliminating costs, reducing dependencies on other companies and partners, or creating its own infrastructure to support new services that demand higher resources.

Importantly, Content Delivery Networks (CDNs) were already emerging players in that scenario. CDNs are networks of servers distributed all over the internet that store copies of highly demanded contents as cache to reduce distance from users to it and facilitate its access.

With CDNs, content access delays are reduced since the physical distance and intermediary points required to access that information is shortened. As content delivery companies, CDNs have been interconnecting non-hierarchically to Internet Service Providers contributing to the formation of a flat network or a “peering mesh,” as Amogh Dhamdhere and Constantine

Dovrolis called it. According to these authors: “The major CDNs and CPs [content providers] have expanded to almost every region of the developed world, so that they can be co-located with many ASes [autonomous systems] at internet exchange points (IXPs)” (Dhamdhere &

Dovrolis, 2010, p. 2). And by escalating their presence, authors note, IXPs have favored peering links and the conditions necessary for the new connectivity model among networks to flourish.

IXPs in the global South14

14 Global South and global North are political categories used in critical studies in humanities and social sciences that go beyond geographic terms and binarism implied by “North” and “South.” They refer to regions’ historical trajectories that encompass politics, economics, and social aspects and go back to colonialism as will be further 38

Significantly, the deployment of IXPs in countries in the global South seems to follow a different path when compared to the global North, primarily studied in the aforementioned research. Recent study on IXP databases suggests that “more affluent regions have a better coverage by IXPs” (Klöti et al., 2016). And while a longitudinal analysis of this unequal scenario is yet to be done, works in Africa have noted out that only a half of the African countries have

IXPs and a great part of them were established recently with the support of volunteer personnel as well as equipment donated by internet international organizations, and they face difficulties to expand (Fanou et al., 2017). Among the challenges there is a lack of incentives for ISPs to interconnect locally at IXPs once the content still comes from other regions, mainly the United

States (Fanou et al., 2016). The benefits of IXPs are generally recognized, particularly in that they allow the joining of many networks at the same place, facilitating private agreements and peering, which may reduce international traffic and traffic costs, and improve, as a consequence, the internet quality for end users by keeping local content locally, diminishing latency and leveraging speed (Fanou et al., 2017). Countries devoid of IXPs are obviously unable to experience such benefits. Nevertheless, in a context where the most accessed content is not local and where the peering dynamics do not work ideally due to companies’ market strategies, even in the presence of a local IXP, network operators tend to not see such benefits either.

Data on global South countries’ Domain Name System traffic is a good indicator of this unequal scenario, especially when compared to the global North. According to Anne Edmundson and colleagues’ experiment focusing on resolution of the most popular domain names15 in Brazil,

discussed. Unlike critical studies, engineering studies use country and continent as analytical categories only. Here and beyond, the use of global South and global North is my categorization. 15 Domain names, popularly known as websites addresses, are alphanumeric ways to represent numeric-only Internet Protocol (IP) addresses on the internet. IP addresses are unique numbers used “to identify the senders and receivers 39

Kenya, India, Netherlands and the United States, authors have shown similar patterns between the first three countries and the last two. Considering where the paths terminate, or where the website content searched is found, only 17% of the Brazilian paths, 5% of the Indian paths and

2% of the Kenyan paths terminate in their own countries, while 77%, 44% and 63% of them, respectively, terminate in the U.S. Further, 0% of the paths from Brazil, India and Kenya terminate in any of these global South countries, and although their geographical distance help to justify that, other elements indicate that content is commonly hosted in the global North, and more specifically in the United States, even when countries are far from there. For instance, over

50% of the most popular domain names in Brazil are hosted in the U.S. In the opposite direction,

97% of the American paths and 39% of the Dutch paths end in their own countries while none paths from there terminate in the global South. In fact, 45% of the Dutch paths end also in the

U.S. while the opposite occurs only in 2% of the paths. The search for content clearly seems to be unidirectional, from the South and the North to the North, where the U.S. is predominant.

The U.S. is also preponderant among the countries that the paths traverse, independent of where the paths end. In this case, 84%, 72% and 62% respectively of the Brazilian, Indian and

Kenyan paths pass through the U.S.; 20% and 50% respectively of the Indian and Kenyan paths also traverse Great Britain. As the authors point out, “Many paths likely traverse Great Britain and the Netherlands due to the presence of large internet exchange points” (Edmundson et al.,

2016, p. 6), where not only content is usually stored in CDNs, but also where numerous networks are reachable. And while a significant number of the Dutch paths (58%) traverse the U.S., none

of packets” (Denning & Martell, 2015, p. 11). When a computer tries to access a website, a request to find the website alphanumeric address and match it to the IP address is resolved, the website is then located and its content, hosted anywhere in the network, can be received at the computer. 40

of the paths from the Netherlands or the U.S. traverse any global South country. This study indicates a common pattern in which content consumed in countries from the South is received from countries from the North, while the opposite is not true, with the consequent international traffic costs that accrue from that.

With a policy focus, engineering studies have tried to recommend some alternatives to reverse the current situation specifically in Africa, where despite the growth of internet access, the internet infrastructure is yet underdeveloped and the prices are still high due to the international traffic (Fanou, Francois, et al., 2017; Fanou, Valera, et al., 2017; Fanou et al.,

2016). One of the possible solutions that scholars propose is interconnecting the IXPs throughout the continent. This would optimize resources, including attracting CDNs currently based in other regions, mainly in the global North, to a single point in Africa. In this way, popular content would be closer to African users. Clearly, this is a moment when social and political variables gain relevance in the technical realm. Among the challenges that the authors identify to such policy implementation is the fact that more than 32% of the IXPs in Africa are based in

“countries that experienced political instability (e.g., Ivory Coast, Egypt, Burkina Faso), rebellions (e.g., DR Congo, Nigeria, Burundi), or terrorist attacks (e.g., Sudan, Nigeria) over the last 5 years” (Fanou, Valera, et al., 2017, p. 5). Risks of outage in these countries, along with lack of optical fiber, are some of the barriers to overcome. With regard to fiber, there are not only economic constraints to make it possible. “For example, DR Congo and Congo do not agree to let any fiber cross their common border” (Fanou, Valera, et al., 2017, p. 5). The politics surrounding sovereignty affect the internet infrastructure design and make some technical attempts unlikely to be realized.

41

2.2. IXPs: Centralizing Points by Design

In contrast to the architecture that preceded the internet built on a centralized and hierarchical telecommunications infrastructure, IXP mesh configurations can be considered radically decentralized. Rather than having a hierarchical switching system with a small number of centralized nodes under the ownership of one organization, the architecture of IXPs distributes these interconnection nodes around the world, albeit imperfectly, and under different governance models and administrators. Despite this fact, IXPs are nodes within the internet distributed ecosystem, and end up being centralizing points by design.

Remarkably, research has found out that through only the five largest IXPs in the world, it is possible to reach 40% of the internet client base, measured in terms of announced Internet

Protocol addresses that the networks interconnected to the IXPs may reach—more than one billion IPv416 addresses in this case. Furthermore, in the case that it is considered not only the networks directly connected, but also one hop—one crossed network in the path—, 91% of the

IPv4 addresses can be reached within these five IXPs (Kotronis et al., 2016).

That can be explained by the fact that as IXPs reach higher connectivity, meaning a greater number of autonomous systems connected, IXPs’ centralization effects increase. This happens because the more participants an IXP has, the more value to new participants it adds.

Such behavior is known as “network effects” or “network externalities,” when “the value of membership to one user is positively affected when another user joins and enlarges the network”

16 Internet Protocol version 4 (IPv4) is the most used and disseminated version of this protocol. The networking community has tried to replace it for the sixth version (IPv6) led by the argument of address scarcity. A critical historical approach of this process can be seen in DeNardis (2009). 42

(Katz & Shapiro, 1994, p. 94)—where user, in the IXP context, means autonomous systems. As

Bernhard Ager and colleagues point out, “ [L]arge IXPs are a microcosm of the Internet as a whole in terms of types of networks, business relationships, or traffic” (Ager et al., 2012, p. 164).

Centralizing technologies and artifacts that concentrate resources are thus not incompatible with the distributed design of the internet. Indeed, IXPs’ centralized design not only contributes to them being recognized as convenient spaces to inform routing policies (Perez et al., 2016) and to be sources for BGP routes collection to check any incidents with the paths that have been propagated (Mathew, 2014); such design has also inspired new services.

2.2.1. Centralization Values in Market Trends

Within scholarship that points to the deployment of new services, research has suggested that IXPs are the most appropriate place to improve the quality of service (QoS) of routing activities and add some guarantees of end-to-end communication by “applying interdomain control over how parts of Internet traffic are routed.” (Kotronis et al., 2016, p. 2). Such control is absent in BGP17, the protocol used to inter-network communication that is considered vulnerable in different ways. As will be further addressed ahead, there are many ways through which internet traffic can be disrupted through BGP, from system misconfiguration to malicious exploitation of protocol vulnerabilities that generate steal of traffic from one network to another

(Murphy, 2006). In view of that, in the new service proposed by Vasileios Kotronis and colleagues, routing would be a kind of ready-to-use service offered to autonomous systems at

17 For a discussion on this protocol, see Appendix B. 43

IXPs as “centralized inter-domain [inter-network] routing brokers” (Kotronis et al., 2016), taking advantage of IXPs’ high connectivity and path diversity.

Also, adding to the general dissatisfaction among the networking community with the low level of control allowed by BGP design, Arpit Gupta and colleagues regret that in using

BGP, “[n]etworks cannot make more finegrained decisions based on the type of application or the sender (…)” when delivering traffic; “[they] have little control over end-to-end paths [;] and

“cannot directly express preferred inbound and outbound paths.” (Gupta et al., 2014, p. 551). To expand the decision capacity of the network operators, they suggest developing a “Software

Defined Internet Exchange,” which would enable application-specific peering, in other words, the possibility of managing peering based on the kind of content that will be exchanged. For instance, “[a]n ISP could configure its edge routers to make different forwarding decisions for different application packet classifiers (to identify the relevant traffic) and policy-based routing

(to direct that traffic over a special path).” (Gupta et al., 2014, p.552). Networks could also increase the control of inbound traffic engineering, meaning control over how traffic from other networks enters their own network “according to source IP addresses or port numbers” (Gupta et al., 2014, p. 552).18

In another work, similarly concerned with internet traffic costs and optimization at a protocol level, scholars have studied routing data available at IXPs to subsidize connectivity decisions of internet applications in the internet infrastructure (Perez et al., 2016). With a focus on the Application-Layer Traffic Optimization (ALTO), a proposed protocol at the IETF, these

18 Port numbers are considered an element of the addressing information in computer networking. They are software-based—distinct from physical devices ports—and allow identification of applications, origin and destination of messages, etc. (Mitchell, 2019). 44

researchers propose a platform built on BGP information from the Brazilian public IXP ecosystem, in which network information is made available to application operators in order to facilitate their traffic decisions. They calculate the costs to transmit data between two endpoints on the internet, which are measured in terms of autonomous systems hops, or the distance between source and destination.

These works aiming to support “traffic engineering”—a “collection of management decisions an ISP makes, including the local configuration of the way the Internet routing protocol (BGP) works, to allocate traffic to the different paths they control.” (Faratin et al., 2008, p. 65), may be read in the networking research community as a response to demands for innovation, security and traffic optimization in routing protocols that have already been voiced in the scholarship (McKeown et al., 2008; Siddiqui et al., 2015). Nonetheless, here they work as exemplar evidence of the centralized design that IXPs have in the internet infrastructure and the kind of governance applications and tracking mechanisms over data packets’ identifiers that such design may enable.

2.2.2. Centralization Values in Core IXP Services

Whereas the examples just analyzed are presented as alternatives and market trends, current services commonly offered by IXPs reveal the centralization values embedded in these internet nodes’ architectures already. Philipp Richter and colleagues note that an important aspect responsible for the growth of IXPs worldwide is the provision of a route server to facilitate peering arrangements and operators’ routing task. In a nutshell, route servers allow that instead of autonomous systems’ network operators setting up multiple sessions with each of the 45

IXPs’ participants that they want to peer, they set up only one BGP peering session with the IXP route server to access the routes that other members connected to it advertise (Richter et al.,

2014). In this case, an autonomous system will have access to a variety of routing options to better decide where to forward its data packets with optimized operation. The authors point out that in their study, the prefixes advertised via the route servers covered more than 80% of the traffic (Richter et al., 2014), meaning that the major part of the addresses reachable through that

IXPs are available through the route servers, showing the prominence of such service.

Route servers facilitate enormously multilateral peering and router communication, adding an important level of concentration at the IXP architecture. The modality of peering, if multilateral or bilateral, is guided by autonomous systems’ peering policies, commonly classified as one of three types: “open policy,” when networks are willing to peer with any other network with no limitations; “selective policy,” when networks follow criteria to peer, such as traffic ratio of the other network; and “restrictive policy,” when networks are closed to new peering (Lodhi et al., 2014; Putta Venkata & Ruan, 2016). Such categorization is default at PeeringDB, a public dataset where autonomous systems voluntarily make information available to facilitate peering agreements.

The autonomous system operators’ interests guide the peering policies that an autonomous system will assume. Big content providers may opt for having open peering policy with small peers, with whom they exchange low traffic at IXPs, and private peering with bigger peers, with whom they exchange high traffic (Lodhi et al., 2014). Thus, it should not be

46

counterintuitive that enormous amounts of traffic on the internet tend to be exchanged through bilateral peerings, which can be supported by IXPs and by private facilities.19

In the peering arena, the network community sees IXPs as facilitators by providing secure locations, expertise and reliable services. Conducting in-depth interviews, Uta Meier-Hahn identifies IXPs’ efforts to be considered “neutral” before their members, and to not be identified as “intermediaries,” given that they do not take part in the peering negotiations (Meier-Hahn,

2014a). Assuming that IXPs may have coordinating roles, she defines them as “organisers,” whose tasks are “aimed at facilitating compromises” (Meier-Hahn, 2014a, p. 18), dissociated from “mediators.” While calling attention to the organizing role that IXPs have is an original approach that adds to the understanding of IXPs, it is fundamental to emphasize that IXPs’ design implicates a core intermediary role in autonomous systems’ relations, impacting outcomes even if they are peering promoters only, and not decision takers.

Specifically, by being a meeting point for numerous networks, IXPs have access to a huge amount of data that, even if not managed by IXP operators, following good practices guidelines, the possibility for accessing it still exists. In a document named “Internet Exchange

Point ‘Wishlist’,” representatives from European IXPs are among the authors who explain the need of having access to members’ data sometimes. As they put it, “It is sometimes necessary to mirror participants’ ports, either because a participant is suspected of some inappropriate activity, to help obtain information when debugging a problem, or during the installation/turn-up

19 On this topic, researchers have found that “high traffic networks are more interested in private peering” (Putta Venkata & Ruan, 2016, p. 6), indicating that big corporations seem to have preferences to exchange traffic in private facilities through bilateral agreements. 47

phase”20 (Hughes et al., 2013, p. 7). In another document, the Brazilian Internet Exchange

Project, a public ecosystem known as IX.br, states in its member policies that:

To ensure that all of this infrastructure operates without setbacks, restrictions are imposed on the type of traffic allowed and authority given to IX.br Network Operation Center (NOC) to temporarily disable Participants [IXP members] who do not comply with the rules or cause disturbances in the operation of the network. (NIC.br & IX.br, 2017, p. 1)

Clearly, mirroring a port to inspect the traffic that is traversing it implies a great amount of power that is delegated by trust to an intermediary player.

Another example of this power is the fact that IXPs attract a great number of routes, although this is not uniform among them. In fact, IXPs in Europe and the United States attract the greatest number of route paths (Nomikos & Dimitropoulos, 2016), which means that a considerable part of the global internet traffic passes through the global North IXPs. Among the top 10 are the cities Amsterdam, in the Netherlands; London, in England; Frankfurt, in Germany,

Palo Alto, Chicago, Ashburn, Miami, Los Angeles and other California cities in the United

States; Stockholm, in Sweden (Nomikos & Dimitropoulos, 2016). Such information is consistent with previous research that shows that the international path routes go mainly through the United

States, Germany and Great Britain (Karlin et al., 2009). Due to this fact, Josh Karlin, Stephanie

Forrest and Jennifer Rexford argue that any routing policies in the networks within these countries tend to have a big impact in the internet routing ecosystem (Karlin et al., 2009).

One of the reasons for IXPs to attract paths is not only their high connectivity due to high number of ISPs attached to them, but also the presence of highly accessed content providers and

20 A port may be logical—software-based—referring to the addressing information in computer networking where through port numbers it is possible to identify applications, origin and destination of messages, etc. Another kind of port is that of a physical entrance into devices that allows the connection of a cable and the formation of a wired network (Mitchell, 2019a, 2019b). 48

CDNs. Similar to router servers, interviewees recognize that CDNs are one of the main reasons for Internet Service Providers to connect to IXPs. CDNs have become key players on the internet since the 2000s, attracting big revenues and being responsible for a great amount of traffic

(Gourdin et al., 2017). In the literature, one finds classifications such as “pure-plays CDNs” (Lee et al., 2014) or “traditional CDNs” (Yu et al., 2012), referring to big players specialized in this kind of service (e.g. Akamai), and also “telco-CDNs” (Lee et al., 2014) or “ISP-CDNs” (Yu et al., 2012), which offer content delivery services as a complement to their functions as ISPs— including Tier-1 ISPs such as Level 3, one of the biggest transit providers in the market that has seen the demand for transit decrease and has invested in CDN business as well (Gourdin et al.,

2017). Additionally, content providers’ CDNs are also important players in this scenario— especially giant companies such as Google, Netflix, etc. have their own CDNs, and some of them, such as Google, also run their own backbone networks and data centers (Böttger et al.,

2016).

These different kinds of businesses pursue different market strategies. For instance, while some ISP-CDNs adopt a more “centralized approach by building a few data centers,” traditional

CDNs tend to adopt a “more distributed approach by deploying its servers at thousands of locations across the world” (Yu et al., 2012, p. 2, emphasis in the original). Due to its extensive scope worldwide, IXPs become strategic hosts for supporting a distributed approach, but as already seen, where CDNs will install their servers and how distributed that will be depend on their own policies and negotiations.

Research on Netflix strategies, a video streaming giant responsible for 15% of the total downstream traffic on the internet (Sandvine, 2018) as of 2018, shows how the IXP ecosystem

49

worldwide helps the company to distribute its content by concentrating its servers at IXPs. From

8,492 Netflix servers identified in the study and distributed around the world, 51% are hosted at

IXPs (Böttger et al., 2016). Nevertheless, when comparing the number of IXP locations versus the number of ISP locations where the servers are hosted, the difference is clear: 52 IXPs versus

569 ISPs (Böttger et al., 2016). Interestingly, in terms of traffic, the study shows that IXP-located servers are responsible for delivering approximately three times more Netflix traffic than ISP servers (Böttger et al., 2016). The authors highlight Netflix preferences for installing its servers in the largest IXPs—the ones with a larger number of participants known to be concentrated in the most affluent areas of the world, with few exceptions (e.g. Brazil). The authors also suggest that the ISPs’ hosts are a geographical complement to IXPs’ hosts, being smaller in size than the

IXPs. This can be illustrated by situations where countries have Internet Service Providers that do not accept to peer at IXPs, incumbents included. As they have a good portion of consumers that Netflix needs to reach, installing servers at ISPs becomes necessary.

While ISPs and IXPs have an interest in having big content provider CDNs inside their network, the cost for maintaining such servers may be a limitation for small ISPs and small IXPs as well. According to Michuki Mwangi, Regional Development Manager for Africa at the

Internet Society in Kenya, when caching content at IXPs, the cost of paying transit providers for keeping the servers updated is a significant amount for IXPs that are commonly built and maintained with international organization support, as occurs in Africa. He mentions that shared cost models, as the Africa IXP Association (AFIX) suggests, end up being a possible response to such difficulties. Along with the infrastructure deficits already mentioned in Africa, it should not

50

be a surprise that when compared to other continents, the fraction of Netflix servers and traffic in

Africa is almost null (Böttger et al., 2016).21

IXPs’ centralizing technical design associated with geographic concentrations raises questions about national states, sovereignty and jurisdiction on the internet. It is at this point that the technical guidelines of internet architecture meet the politics of interconnection, where data packets need to be treated not as amorphous bits but content, as well as national borders need to be not marginal to explanations but essential variables to understand the internet exchange points ecosystem. And when the concept of data becomes associated with access, meaning and regulation, conditions are set to conceive data as “information.”

According to Amelia Acker, “the difference between information and data is context. A piece of data without context is without meaning, but when data is put into context through practices such as aggregation, description, classification, organization, or application, it becomes meaningful information to people and machines.” (Acker, 2015, p. 1-2). Using Vint Cerf’s analogy of data packets as postcards that have a to/from field and another area for the message

(Denning & Martell, 2015, p. 227), it is possible to notice that there is a distinction between data and information implied in the engineering inside explanations that authorizes them to describe the transportation process on the internet as data transportation only. Nevertheless, what is being transported above all is potential information, with fields that allow to make sense of the context.

It is the transportation of potential information passing through distinct nodes that concerns this dissertation.

21 For the sake of comparison, this scenario is completely different when considering big ISPs. Due to their market power, they may push big content providers to be their clients, instead of peering with them or hosting their CDNs. This is the case of Netflix with AT&T, Comcast, Time Warner Cable and Verizon in the United States and Deutsche Telekom in Germany (Böttger et al., 2016). 51

3. Theoretical Literature and Conceptual Framework

The study of a technological artifact such as internet exchange points does not come without challenges. First, they are not visible to the general public and have complex technical characteristics.22 Second, they have been increasingly associated with broad narratives of development, due in part to the way that international organizations advocate for their creation to leverage broadband deployment and quality (Agudelo et al., 2014; Blackman & Srivastava,

2011; Intven et al., 2000; Weller & Woodcock, 2013), and these imagined and desired IXPs’ affordances blur other artifact attributes. Third, IXP policies are not public as one could expect

(or hope for), and accessing them depends strongly on the openness of the IXPs’ administrators.

These organizations can be private, not-for-profit, governmental or multi-stakeholder, and can be subject to organizational context, and culture, among other variables that characterize the governance of large technical systems (Ruano-Borbalan, 2017). Fourth, the unequal distribution of IXPs between the global North and the global South adds political and economic layers to the task of de-black-boxing their attributes. When closed, “A black box contains that which no longer needs to be reconsidered (…)” (Callon & Latour, 1981, p. 285). The study of IXPs require to unfold some elements to be further investigated.

Beyond the numerous challenges intrinsic to the subject of study, there is a specter that haunts social scientists who aim to focus on technologies themselves: “technological determinism,” or the idea that technology is a result of either its own paths or generic advances

22 In common usage, technical is an adjective used in opposition to political. This is not the meaning adopted here, where the term refers to indispensable and highly specialized functions that are invisible to the general public (Latour, 1999). 52

in science, yet has the intrinsic ability to impose significant effects on society (MacKenzie &

Wajcman, 1999). W. Brian Arthur (2009) provides a good example of this approach. An economist influenced by Darwinism, Arthur defends that “technology creates itself out of itself”

(Arthur, 2009, p. 11). Disregarding their social construction, this author understands that by looking at the “inside anatomies” of technologies, it is possible to draw a theory of their foundation and evolution (p. 5). This is the kind of view that may have consequences on how individuals will react to changes and implications of technology in society, encouraging people’s adaptation instead of propensity to shape these processes (MacKenzie & Wajcman, 1999).

Unlike Arthur, in Science and Technology Studies (STS), Langdon Winner states that technology cannot be understood as “the sole result of an internal dynamic and then, unmediated by any other influence” (Winner, 1986, p. 2). Nonetheless, he has received criticism for his technological deterministic views (Joerges, 1999; Lievrouw, 2014), mainly because of the prominence of predictable social effects of technology in his work, including the concept of

“autonomous technology” (Winner, 1978) and the focus on consequences of the design of a bridge to people’s mobility (Winner, 1980). Winner is one of the critics of social constructivism in the STS field, and the imperative of social patterns and actors’ dynamics in the study of technologies, defending that technologies themselves have material consequences and social effects that need to be examined (Lievrouw, 2014; Winner, 1978, 1980).

Langdon Winner’s work responds to social constructivists who, for instance, emphasize that “The social environment (…) shapes the technical characteristics of the artifact” (Bijker,

Hughes, & Pinch, 1987a, p. 12). Such scholars defend the importance of studying technologies beyond their usage, including their design and technical matters, and highlight the social shaping

53

in any of these layers. Building what they call the social construction of technology (SCOT) approach, they distinguish three stages of a technology, which are adapted from the sociology of knowledge: a) “interpretative flexibility,” which refers to the fact that a technology design can be open to different interpretations according to the social groups involved; b) mechanisms of

“stabilization” and “closure” of a technology that suspend the interpretative flexibility; and c) the ways that stabilization and closure occur in face of a wider social, political and cultural context

(Bijker, Hughes, & Pinch, 1987b, pp. 40-46). If Wiebe Bijker, Thomas Hughes, and Trevor

Pinch could be criticized by Langdon Winner for having a social deterministic approach, there is an underlying level of this approach related to where the authors are situated and their own social bias. This is evident when they apply their framework to the development of bicycles. They quote a review of an annual exhibition from the nineteenth century followed by a conclusion that renders their perspective explicit:

From the number of safeties adapted for the use of ladies, it seems as if bicycling was becoming popular with the weaker sex [sic], and we are not surprised at it, considering the saving of power derived from the use of a machine having only one slack. (Stanley Exhibition of Cycles, 1890, pp. 107-108) Thus some parts of the bicycle's development can be better explained by including a separate social group of feminine cycle-users. This need not, of course, be so in other cases: For instance, we would not expect it to be useful to consider a separate social group of women users of, say, fluorescent lamps. (Bijker, Hughes, & Pinch, 1987b, p. 34, emphasis and "sic" added)

The three authors want to illustrate above how a “feminine” narrative somehow shaped the development of the bicycle, which means for the authors both the consideration of a “social group of feminine cycle-users” in the process, and an evidence of how different social groups may interpret a technology differently—an example of the interpretative flexibility. Safety reasoning appears an important factor of stabilization and closure; however, the authors fail in applying the third stage of their own approach, which is considering a wider social context to 54

analyze what safety means: a masculine expectation of women’s capacity and behavior. There are two points that are worthy to highlight in their analysis.

First, the social constructivist view exemplified by these authors does not include a critical view on what “social” may signify. By focusing on the “protagonists” that shape the controversies related to the development of technologies without problematizing who they are, their analysis fails in making explicit the values embedded in the bicycle’s deployment in direct relation with the “social-cultural milieu” (Bijker et al., 1987b, p. 27), as the authors call it.

Secondly, groups that are supposed to be influencers, because they are mentioned as a focus of the process, may end up not being de facto influencers. In the case above, beyond 19th century men interpreting what it was to be a woman at that time – “weak,” fragile –, the authors do not bring any evidences that, for instance, feminine engineers were influencing the bicycle design to respond to their gender needs. The male trio fails to interpret the “multidirectional flux” (Bijker et al., 1987b, p. 13), as they name it, of the societal phenomenon that they examine: they interpret men’s feminine imaginary as feminine influence.

This kind of bias does not seem to be localized. It has been already observed in the literature that “‘social shaping’ does not necessarily involve reference to wider societal relations such as those of class, gender and ethnicity” (MacKenzie & Wajcman, 1999); however, there are useful exceptions that call attention to the social values that embed technologists and designers in general. For the feminist and Marxist Cynthia Cockburn, this is why “Industrial, commercial, military technologies, are masculine in a very historical and material sense. They cannot readily be used in a feminine, nor even a sexless, mode” (Cockburn, 1983).

55

Donald MacKenzie and Judy Wajcman argue that one of the most important contributions of Cockburn is that her work implicates that not only technology is shaped by gender, as social constructivists would argue, but gender is also shaped by technology, given that women have been systematically kept apart of engineering areas while technology “is one of the formative processes of men” (Cockburn, 1983, p. 17). In the bicycle case, where the authors identify that the “dressing problem” (Bijker et al., 1987b, p. 38) had to be solved to include women in their long skirts as cycle users, it should not go unnoticed how technology designed by men was shaping society at that moment, keeping moral and etiquette issues intact. In this scenario, “technology is society made durable” (Latour, 1991), not only in terms of the different temporalities that it embeds—as Latour intends to mean here—but in the sense that it contributes to the permanence of patterns of relations. This is an important rationale to think of the role of

IXPs as a technology that although emerges with the internet and its distributed architecture, keeps centralizing characteristics that, as we shall see, will contribute to the continuation of social, political and economic phenomena.

As this section has sought to show, authors have recognized the importance of studying technology’s materiality. And some credit concurrent streams of thought for their contributions:

“Technological determinism contains a partial truth. Technology matters.” (MacKenzie &

Wajcman, 1999, p. 2). Nevertheless, social factors such as culture, politics and economics cannot be disregarded (MacKenzie & Wajcman, 1999).

In studying internet interconnection infrastructure, it is therefore appropriate to look at design and governance, technology and society as interlaced. Let’s understand how this

56

discussion has influenced the research approaches on communication technologies, specifically, the internet.

3.1. Communication and Internet Governance Research

Communication technologies are considered to be composed by three intertwined elements: artifacts, practices and social arrangements, where practices are related to individuals’ actions to produce or use such artifacts, and social arrangements include the artifact’s governance and organizations that emerge around the artifacts and practices (Lievrouw &

Livingstone, 2010b, 2010a). Still, in communication and media studies, the study of the technological artifacts has been preempted in favor of other focuses. As Leah Lievrouw argues,

“communication research seems to be slower to adopt explicitly materialist accounts of technology, an idealist tendency that has been reinforced by the embrace of social constructivism from STS and critical/cultural media studies” (Lievrouw, 2014, p. 44).

An example of that is the seminal and influential work of John Carey “Communication as

Culture: Essays on Media and Society” where defending “technology as culture” he suggests that

“The first task is to demythologize the rhetoric of the electronic sublime. Electronics is neither the arrival of apocalypse nor the dispensation of grace. Technology is technology; it is a means for communication and transportation over space, and nothing more” (Carey, 2008, p. 107).

Lievrouw explains that the predominance of culturalist approaches in which technologies are seen as a product of culture only prevent communication research to theoretically formulate on the “the intrinsic social and material character of communication technology as equally definitive and co-determining” (Lievrouw, 2014, p. 24), accounting for a parallel technology role on 57

societal phenomenon. Interestingly, the name of the book in which her article figures is “Media technologies: essays on communication, materiality, and society,” intentionally or not remembering Careys’ book title.

This trend in the field contributes to the choice of some objects of study instead of others.

Scholars have noted that research on the World Wide Web (WWW), focused on content and internet use, has represented much of the studies in this area (DeNardis & Hackl, 2015;

DeNardis, 2010). Despite that, there is an emerging scholarship that has contributed to the study of artifacts, with researchers interestingly engaged in discussions of gender, ethnicity and class associated with technology design, and not with users’ practices as is more common. Examples include findings on the differences of ads delivered on search engines to racially-associated names (Sweeney, 2013), the surly reception of feminist discussions in online platforms as

Reddit.com (Massanari, 2017), the consequences of social media privacy design to the creation of a privacy divide in society (Papacharissi, 2010), and yet the way that internet governance domains serve as gateways of LGBT (Lesbian, Gay, Bisexual, and Transgender) discourse

(DeNardis & Hackl, 2016).

In fact, an important subject area that has been continuously compromised with understanding artifacts is the field of internet governance, specifically in what concerns internet architecture, internet platforms and their design associated with social, political and economic factors. Importantly, within this scholarship, internet design and governance are considered two sides of the same phenomenon (DeNardis, 2014; Musiani, 2013), a fact that results in approaches that take into consideration the social arrangements in which technological artifacts are involved, including, on the one hand, the institutions and the modes of governance and regulation where

58

they are inscribed, and on the other, the kinds of regulation and governance that they lead to. As an example of the latter, Laura DeNardis and Andrea Hackl state that their investigation of social media platforms examines “governance by social media rather than governance of social media”

(DeNardis & Hackl, 2015, p. 2, emphasis in the original), bringing artifacts up front.

An important contribution of this approach is to illuminate the social, political and economic aspects that emerge from unnoticed arenas of dispute and policy decision taken by non-traditional policy actors. This extends to infrastructure layers more distant from users’ practices, where apparently drier and overly technical communication research joins efforts to leverage public understanding of technology by unveiling knowledge from its core. Indeed, as

Laura DeNardis puts it, “[The internet] architecture is not external to politics and culture but, rather, deeply embeds the values and policy decisions that ultimately structure how we access information, how innovation will proceed, and how we exercise individual freedom online”

(DeNardis, 2010, p. 1).

Nevertheless, internet governance as a subject area is by no means uniform. In fact, its transdisciplinary characteristic is shaped by dialogues between communication, computer science, engineering, international affairs, legal studies, political sciences, sociology, science and technology studies, among others. And as Rekha Jain notes, there are areas “that encompass aspects of Internet Governance, [but] do not label themselves as studies of Internet Governance”

(Jain, 2015, p. 4). Scholars who have tried to examine the diverse intellectual sources mobilized in this broad area show how diversely researchers and policymakers understand it (Ziewitz &

Pentzold, 2014). The authors have identified five “modes of ordering,” but unfortunately, they are not clarifying, as they name one of the categories with the same name that they are inquiring

59

(Internet Governance, Cyberlaw, Online self-governance, Governmentality and Surveillance,

Techno-scientific governance). While this kind of effort may contribute to a sense of the distinct narratives about “internet governance” being used, the categories are more likely to appear merged in the works of such a transdisciplinary subject area. In the internet governance scholarship in which the present dissertation relies, legal studies, science and technology studies, infrastructure studies, along with surveillance studies, are brought together mobilizing different perspectives to frame communication research, and providing a substantive scholarship to support the present study of internet interconnection infrastructure.

Laura DeNardis posits that “The primary task of Internet governance involves the design and administration of the technologies necessary to keep the Internet operational and the enactment of substantive policy around these technologies” (DeNardis, 2014, p. 6). In this reference book, she seeks to delimit the field of inquiry to internet-specific topics and technologies, among which she mentions “infrastructure management techniques, interconnection agreements among network operators at Internet exchange points, and the development of standards on which the Internet operates” (DeNardis, 2014, p. 21). In a previous work, she identifies the lack of studies about internet interconnection and its “critical implications.” The author notes that “Interconnection agreements are unseen in that there are no directly relevant statutes, there is no regulatory oversight, and there is little transparency in private contracts and agreements” (DeNardis, 2010, p. 12).

In fact, as networks are autonomous, corporations who run them may decide to not connect to specific networks, claiming business reasons, such as competition issues, disadvantages regarding infrastructure and traffic sizes, etc. Examples of that have already

60

occurred between two big Internet Service Providers in the United States: Cogent and Sprint. In

2008, these networks stopped interconnection due to a business disagreement about setting peering or transit arrangements, which left many users from both networks uncommunicable

(Weiser, 2009). In another case, in 2007, Comcast deliberately ended connection with

BitTorrent, a content provider responsible for huge amounts of peer-to-peer data transfer among users, claiming network management reasons (Weiser, 2009).

While this kind of events has been framed by some as a type of internet fragmentation issue (Hill, 2012), it shows the preponderant role of private actors and their technologies in this realm. Moreover, it provides elements to understand what Laura DeNardis calls “privatized governance,” and the fact that private actors, including companies and nongovernmental institutions, coordinate and run most of the internet functions de facto, having direct, although usually disregarded and non-problematized, effects on individuals’ access to knowledge, freedom of speech and privacy—recurring communication rights advocated in this literature originated in the United States’ political and societal contexts. Moreover, these companies have their origins in some nations (e.g. United States and Western countries), more than in others

(global South), which constitutes a fundamental background to the study of internet exchange points designed to facilitate the interconnection of these companies.

The consequent challenges accrued from this scenario are first a question of information intermediaries’ legitimacy—within their countries and abroad—to exercise the role of mediators of rights, a role historically exercised exclusively by governments. Within the content layer, the private regulation of people’s rights occurs in view of a shift to self-governance, in which content platforms rely commonly on (hardly read but) accepted consent terms opening the door

61

to algorithm governance through specific platforms design. Whereas scholarship has critically discussed content providers’ policies, premises and positions (DeNardis & Hackl, 2015;

Gillespie, 2010; Sargsyan, 2016), the privatized governance has its scope deeply amplified when other infrastructure intermediaries, commonly from a reduced number of countries from the

North, are also taken into consideration. For instance, the dynamics related to networks’ decisions or IXPs’ influence on decisions about users’ data routes and flow in the internet occur in highly unknown territory and similarly affect communication rights, including through jurisdiction issues.

Specifically from legal studies comes the discussion on “extraterritorial extension of sovereignty” (La Chappelle & Fehlinger, 2016) of some nations. Because of the prominence of some countries’ IXPs, which attract more paths and more content, as previously shown, their national jurisdiction, including local courts’ decisions, can have even more direct effects on how data of citizens from other countries will be treated. As Jennifer Daskal points out, “whereas data transits the globe without any intrinsic ties to territory, the governments that seek to access or regulate this data operate with territorial-based limits” (Daskal, 2018, p. 180). Still, as the previous section on the architectures of IXPs demonstrates, autonomous systems are “sovereign” in their decisions—a term used by the network specialist Adilson Florentino during his interview with the author. Thus, national states’ sovereignty and privatization of governance led by

“sovereign” corporations form a complex topic in internet governance scholarship.

This becomes even more intricate in view of what Laura DeNardis has called “the turn to infrastructure for content control.” With an initial focus on the risks for access to knowledge and freedom of speech (DeNardis, 2012), later expanded to other threats such as surveillance

62

(Musiani et al., 2016), the author argues that social and political conflicts have led to motivated communication technologies interventions, showing that internet architecture has been used as

“proxy” for governments to control information (DeNardis, 2012, p. 721).

3.1.1. Asymmetries in Internet Governance Seen from the South

Within internet governance practice and research, the relationships between government and internet corporations are interpreted by policymakers from non-Western countries as being scheming, marked by a Machiavellian synergy that strengthens powerful national states as the

United States (Jain, 2015; Nocetti, 2015), where internet giants are originally from. The reasons to resonate this rationale were intensified after Edward Snowden’s revelations in 2013 in which he leaked confidential U.S. government documents stating collaborations between intelligence agencies, as the National Security Agency (NSA), and big U.S. corporations. For instance, in documents of the NSA PRISM operation, in which data collection from companies’ servers without judicial order has been revealed, one can read things like “MS [Microsoft], working with the FBI [Federal Bureau of Investigation], developed a surveillance capability to deal with the new SSL [Secure Sockets Layer]. These solutions were successfully tested and went live 12 Dec

2012” (Greenwald, 2015, p. 115). In turn, in the FAIRVIEW operation documents, focused on obtaining aliens’ data, bullet-type sentences introduce the relationship between a non-disclosed corporation and the U.S. government:

Corp partner since 1985 with access to int. cables, routers, switches. The partner operates in the U.S., but has access to information that transits the nation and through its corporate relationships provide unique accesses to other telecoms and ISPs. Aggressively involved in shaping traffic to run signals of interest past our monitors. (Greenwald, 2015, p. 105).

63

Grounded on these facts, accusations of opportunistic United States’ positions on internet governance issues are recurrent. As Julien Nocetti puts it:

Indeed, Russia, like China and some Middle Eastern nations (particularly in the Gulf ), considers the US stance on cyber politics to be largely hypocritical: while preaching the tearing down of ‘digital borders’ that have emerged in some authoritarian countries, US intelligence organizations have been recording and exploiting mass surveillance data— without any control, thus undermining Washington’s ‘ability to act hypocritically’ (Nocetti, 2015, p. 120-121).

Such arguments should not prevent us from considering that other countries, including

U.S. critics in the global South, can try to use its same formulas, strengthen their own companies and conduct surveillance as well. For instance, there has been evidence reported in the news that

Vodafone, a European telco company, had in 2009-2011 identified back doors installed in

Huawei’s home internet router software, which could have generated unauthorized access to networks in Italy (Lepido, 2019). And the relations between Huawei and the Chinese state is an open question—recent studies from Fulbright University Vietnam and the George Washington

University have even suggested that Huawei may be considered a state-owned company (Balding

& Clarke, 2019).

Nevertheless, the relationship between companies and states presents different nuances.

In certain situations, this relationship is not necessarily marked by complicity, as when they are required to respond to governments giving access to their customers’ private content. Laura

DeNardis notes that the kind of task delegation in place between government and internet information intermediaries nowadays is a phenomenon that differs from previous delegations in other businesses. In some cases, information intermediaries may end up being “compelled to carry out law enforcement functions traditionally performed by the state without compensation

64

and often with additional expense and possibly even liability exposure” (DeNardis, 2014, p. 13).

There are multiple interests at stake and contingent conditions surrounding these relations.

On the other hand, the “market for surveillance,” as Ronald J. Deibert and Masashi Crete-

Nishihata from the Citizen Lab in the University of Toronto call it, go beyond delegation of law enforcement functions. It involves multiple companies, with different origins, offering specialized technologies that serve governments’ interests to buy them. For the authors, “in the absence of restraints to the contrary, businesspeople will seek out and exploit commercial opportunities of a growing cyber-arms” (Deibert & Crete-Nishihata, 2012, p. 353).

The prominence of the United States in the internet governance realm deserves consideration in this section. According to Milton Mueller, two events mark the internet governance institutional formation: the creation of The Internet Corporation for Assigned Names and Numbers (ICANN) in 1998, with the goal of having a non-governmental model to the governance of critical internet resources, and the World Summit for the Information Society

(WSIS) in 2003 and 2005, when that governance model was brought to question due to the critique of some governments about the prominence of the United States in the governance of such resources and the request for more internationalization (Mueller, 2010; Rosa & Vicentin,

2016). ICANN is known for what has been called multistakeholder model, where governments are considered one actor to be heard among others such as the private sector, the technical community and the not-for-profit organizations. However, from an Indian perspective, “These

[multistakeholder] spaces have been largely dominated by developed countries, predominantly the USA and western countries” (Jain, 2015, p. 4).

65

Numbers confirm such perception. In a conference on Internet Governance and

Cybersecurity at Columbia University in 2015, Paul Twomey, a former ICANN Chair, showed that 33% of the ICANN’s board was from the U.S. and 27% was from Europe. In turn, India was represented with less than 4% and China with 1% of the board. For a comparison, in 2014, the

U.S. population connected to the internet represented only 9% of the whole internet population, while China’s population represented 21% (Columbia University, 2015, pp. 23-24).

An alternative metric to discuss such imbalance of representation in these organizations refers to another important multistakeholder forum, the IETF, where the technical community defines standards for the internet, from protocols to good practices. The communication in the community occurs through documents known as Requests for Comments (RFCs). According to public data regarding the 10,028 RFCs written as of January 11, 2019, 50.8% of the authors are from the U.S., 19,7% are from Europe, and only 6,6% are from China, the first global South country to be listed. India comes in 6th with 1,6% of the authors so far (Arkko, 2019b). Authors’ affiliations provide important background to this scenario, given that the big majority of participants of internet standard forums are affiliated with companies, and represent their interests in innovation and in keeping their market position (Rosa & Vicentin, 2016, p. 73). For instance, the Huawei’s market position contributes to understanding China’s appearance in the list.

Below is the chart with the number of authors per company at the IETF, featuring mostly

U.S. and global North corporations. The few universities—Massachusetts Institute of

Technology (MIT) and University of California, Los Angeles (UCLA) and Berkeley) and the unique intelligence agency (NSA) that figure in the chart are from the U.S.

66

Figure 1. Number of authors per company at the IETF. Source: Reproduction from Arkko, 2019a.

The discrepancy between United States’ corporations and the rest can be partly explained by the country’s protagonistic role in the beginning of the internet, supervening the strong U.S. government investments in computing during the Cold War, which clearly left Russia—one of today’s discontented voices—behind (Wilson, 2009). As the development of the U.S. technology industry cannot be disconnected from political factors and motivations to strengthen the state at that period, nowadays, government representatives from other countries have continuously questioned the legitimacy of the current internet governance institutions in which U.S. companies lead. For instance, “Critics of the IETF, such as Russia and China, have stressed the fact that Americans have dominated the organization from the beginning, arguing that this makes it nothing more than an instrument of US political and commercial interests” (Nocetti, 2015, p.

123).

These debates over hegemony in the global scenario are key to understanding that to look at the global North and the global South symmetrically in internet governance research, as this dissertation proposes, requires not taking for granted narratives established in the field, including

67

the notions of “multistakeholder” or “multilateral” models of governance, which are differently assimilated depending on where one is situated. As Geoffrey Bowker and Susan Leigh Star point out about categorization processes, “Each standard and each category valorizes some point of view and silences another. This is not inherently a bad thing—indeed it is inescapable. But it is an ethical choice, and as such it is dangerous—not bad, but dangerous” (Bowker & Star, 2000, p.

5-6).

Supporters of the multistakeholder model stress the benefit of distancing internet governance from a government-centered and restrictive approach to give voice to different stakeholders. David Clark’s axiom: “We reject kings, presidents and voting. We believe in rough consensus and running code” (Lessig, 1999, p. 1413) synthesizes the spirit away from governments, at the same time that allows to raise the question about who are “we” for one of the internet founders. Multilateral supporters, in turn, defend that governments have the legitimacy in the public realm to make policy. For them, the multistakeholder model favors the capture of private interests and the U.S. dominance, and as such, raises barriers for the representation of interests from countries with “lower resources” (Jain, 2015, p. 4).

Clearly, it does not help to associate the multistakeholder model with “democratic” countries and the second with “authoritarian” countries, as commonly occurs. Warned against that, in this dissertation there will be an effort to approach controversial topics as “boundary objects,” a term that comes from infrastructure studies and their interplay with science and technology studies.

68

3.2. Infrastructure Studies

Boundary objects is a concept used to explain the possibility of dialogue in heterogeneous fields such as science, where different social worlds coexist and collaborate even without consensus. It emerges in the work of Susan Leigh Star, and one of its most used definitions states that “Boundary objects are objects which are both plastic enough to adapt to local needs and the constraints of the several parties employing them, yet robust enough to maintain a common identity across sites” (Star & Griesemer, 2015, p. 176). In the context of science, the authors show that different groups working on the same topic may agree at a certain extent but end up customizing boundary objects according to their internal group view. A state map, for example, may have different representations in which they have the same borders, but the content is adapted depending on the perspectives of the group managing it to defend a certain thesis (Star & Griesemer, 2015). Boundary objects are then subject to “interpretive flexibility”

(Star, 2010), and express “work arrangements that are at once material and processual” (Star,

2010, p. 604), in other words, the social and material factors that surround it matter.

Concepts, theories, tools and technologies as well as humans can all be boundary objects, and the list can be extended. Scholars have built on the concept, along with Actor Network

Theory, to analyze technology as “boundary objects with agency,” calling attention to three dimensions: (i) the social world of who design and use it, and are responsible for shaping the technology; (ii) the characteristics of technology itself; and (iii) the agency of technology, which also impacts the social worlds in which they are involved (Fleischmann, 2006). For this author, technologies “actively reshape relationships within and among these social worlds, shifting alliances, conflicts, and the overall balance of power” (Fleischmann, 2006, p. 82). 69

From infrastructure studies comes also the discussion on the poetics of infrastructure that adds and is an intrinsic part of the politics of infrastructure, vastly approached in internet governance. Brian Larkin makes the case that beyond its technical characteristics, infrastructure also mobilizes emotions and “operate on the level of fantasy and desire” (Larkin, 2013, p. 333) in dynamics that include government and society.

Finally, another crucial contribution comes from the approach to large technical systems, in which transportation, electricity, telephony have been studied with specific focus on the interlinking systems. Jane Summerton’s scholarship is key in this regard as she analyzes what she calls the “politics of interconnection in sociotechnical systems and networks” (Summerton,

1999, p. 94). By studying alliances between electricity corporations in Europe, the author accounts for two parallel processes: the interconnection in the level of the system, where territorial coupling occurs based on the establishment of technical standards across political borders, and the maintenance of autonomous systems in the administrative level. In her precise analysis of Actor Network Theory, she interprets that “sociotechnical alliances and networks are the expressions of multifaceted strategies by which entrepreneurs of technology seek to expand their power and increase their control over other actors, entities and resources that are crucial to reaching their own goals” (Summerton, 1999, p. 95). In view of the strict economic narratives that companies’ representatives raise as motivations for building the alliances, the author sheds light on the political facets of such arrangements that may be related to amplifying their reach to other territories, growing power in relation to competitors and increasing levels of control. In her words, “the ‘economics’ of technology cannot be interpreted apart from their inherent politics

70

and the role of rhetoric in articulating, sustaining and expanding these politics.” (Summerton,

1999, p. 109).

Together, this scholarship brings significant contributions to the research of internet interconnection, not yet examined with an infrastructure studies lens. As we shall see, multiple desires, expectations and interpretations as well as constant disputes among actor-networks surround the deployment and design exploitation of IXPs for surveillance purposes.

3.3. Surveillance Studies

Along with internet governance and infrastructure studies, surveillance studies is another pillar of the present dissertation. This is a field that develops in tandem with the extraordinary surveillance capabilities surrounding physical and virtual spaces nowadays. A foundational reference is Michel Foucault’s theory developed in Discipline and Punish (1977), where the so- called panoptical surveillance builds on an architecture that supports the institutional gaze of an observer coming from above—as indicated by the prefix “sur-veillance” in French (Bossewitch

& Sinnreich, 2013). The gaze is directed to individuals subjected to self-governance in response to the expectation that surveillance is happening, although the panoptical surveillance infrastructure does not allow anyone to be sure about the presence of someone observing from behind the wall. In this scenario, there is an evident imbalance of power that can be visualized not only through the established information flow, where who surveils collects and holds more information than the one who is surveilled (Bossewitch & Sinnreich, 2013), but also through the characteristics of who surveils and is surveilled: in the traditional panopticon, inmates clearly occupy an unprivileged position before the institution. 71

The imprisonment context in which surveillance studies scholarship has been cultivated has inspired theorization of the current scenario characterized by massive monitoring through state apparatuses and privately governed communication technologies, as progressively unveiled by investigative journalism and leaks (Bamford, 2009; Greenwald, 2014; Klein, 2009; Priest &

Arkin, 2011). For Jessa Lingel and Aram Sinnreich, that brings us to a condition of

“institutionalized incarceration,” where the meaning of being incarcerated goes far beyond the physical confinement in a prison. It refers to social constraints that pervade quotidian life and relationships—a situation that is historically primarily experienced by unprivileged societal groups. The authors defend that institutionalized incarceration can be compared to what happens in a context of digital massive surveillance, in a process that they call “incodification,” where ordinary and intimate relationships are now by default mediated by digital technology “that permits, and even presumes, ubiquitous surveillance” (Lingel & Sinnreich, 2016).

This is in line with what Mark Poster first called “Superpanopticon.” The author, who looked at societal phenomena with focus on discourse and language, would write at the time of dissemination of computers and purchase electronic transactions:

Modern society may be read as a discourse in which nominal freedom of action is canceled by the ubiquitous look of the other. It may be interpreted semiologically as a field of signs in which the metadiscourse of the Panopticon is reimposed everywhere, even in places in which it is not installed (Poster, 1990, p. 91).

Shoshana Zuboff has recently theorized this phenomenon with economic lens, defining surveillance as the core of the current economy in a new era of information capitalism, which she calls “surveillance capitalism.” This “aims to predict and modify human behavior as a means to produce revenue and market control” (Zuboff, 2015, p. 75) in an environment in which companies assume that “habitats inside and outside the human body are saturated with data and

72

produce radically distributed opportunities for observation, interpretation, communication, influence, prediction, and ultimately modification of the totality of action” (Zuboff, 2015, p. 82).

This broader capacity of surveillance nowadays, more horizontal by nature, has also more hybrid and complex modus operandi well captured by the term “surveillance assemblage”

(Haggerty & Ericson, 2003). According to these authors, “This assemblage operates by abstracting human bodies from their territorial settings and separating them into a series of discrete flows. These flows are then reassembled into distinct ‘data doubles’ which can be scrutinized and targeted for intervention” (Haggerty & Ericson, 2003, p. 606).

These are key references to build on as IXPs have become attractive sites for state surveillance.

3.4. Values and Technology Design

A general guideline that permeates this dissertation is the study of values embedded in design. This is a broad and interdisciplinary subject area that conveys computing ethics, human computer interaction, information studies, media studies, participatory design, science and technology studies among other areas, with a wide range of interests and approaches. It includes the investigation of the designers’ values, users’ values, values embedded in technologies features, in policy debate, etc. (Shilton et al., 2014). From this area, inspiration to the present research comes from studies that rely on “technical investigations,” which similar to significant scholarship in internet governance, are interested in understanding the way that system attributes enact or constrain values (Friedman et al., 2006; Shilton et al., 2014). These studies are focused not only on the “potential” but also “performed” values of technologies already built and in use 73

(Shilton et al., 2014, p. 6). Some will argue that they examine “technology itself,” complementing conceptual and empirical investigations, the latter focused on “the individuals, groups, or larger social systems that configure, use, or are otherwise affected by the technology”

(Friedman et al., 2006, p. 4). I understand this assertion as calling attention to the primary focus of research, given that this dissertation will show that technology is implicated in multiple social relations.

Shilton et. al (2014) classify Brey's (2000) work on computing ethics and Nissenbaum's

(2009) work on privacy as technical investigations. For instance, building on Winner (1980) and others, Phillip Brey assumes that computer technologies hold “moral properties” and suggests a

“disclosive computer ethics” approach to unveiling such attributes. In this sense, in a prescriptive paper, the author suggests that a given technology can be studied from the standpoint of a chosen value extensively accepted in society (e.g. justice, privacy, etc.). Following that, in a second stage, a theory-driven analysis could be offered for developing a normative examination, possible policy guidelines, and theoretical generalizations (Brey, 2000).

More recently, values-oriented design research has disputed assumptions of previous scholarship in this area. Researchers acknowledge two main scholarly groups in the study of value-oriented design in the United States, Values in Design (VID) and Value Sensitive Design

(VSD), which have been described by their likenesses in “explor[ing] the ways in which moral or social values become part of technological artifacts” (Shilton et al., 2014, p. 2). VSD has been developed in close relation to design in human computer interaction aimed at influencing design processes. Critical issues regarding such an approach are related to the necessity of transposing understandings of values universalism by adopting pluralistic views, situating who are speaking,

74

and making participants’ voices more explicit in research (Borning & Muller, 2012). These points will be further discussed in the next section, as they are in line with the methodology literature in which this dissertation relies. Developing a sensibility to these issues allows one to see power relations embedded in design as Shaowen Bardzell, a feminist scholar, notices:

A timeless and universal stance in cross-cultural design is dangerous because it demotes cultural, social, regional, and national differences in user experiences and outlooks. It also quietly and usually unintentionally imposes—without transparent or rational justification—Western technological norms and practices (Bardzell, 2010, 1305).

Parallel to feminist scholarship influence on the discussions of technology design, Katie

Shilton points to shifts in the notions of values in the recent literature (Shilton, 2018). That responds to demands originated in the practice of design, where pre-conceived lists or inventories of values, characteristic to VSD approach (Cheng & Fleischmann, 2010; Friedman et al., 2006), have shown limitations because a) they raise barriers to the discovery of local values

(Le Dantec et al., 2009), b) may end up being representations of “typical values from a liberal and relatively privileged Western perspective” (Borning & Muller, 2012, p. 4), or yet c) because such lists become difficult for designers to engage with during design processes as they are disconnected from specific practices and needs (JafariNaimi et al., 2015).

For the purpose of the present dissertation, focused on internet infrastructure as a

“deployment” process, and not just as a “technical design” (Dourish, 2015, p. 200), there are convergences with recent values-oriented design literature in that values are associated with

“processes or practices” (Shilton, 2018, p. 249). While assuming values as “guiding principles of what people consider important in life” (Cheng & Fleischmann, 2010, p. 2), it is useful to think of values in context; as Nassim JafariNaimi and colleagues suggest, “designers employ values to serve situations of practice, rather than applying them from precise definitions or following

75

procedural steps” (JafariNaimi et al., 2015, pp. 99-100, emphasis added). For these authors, values are considered “hypotheses,” and as so, “they cannot be applied fully; rather, they serve in that they help us make sense of new conditions” (JafariNaimi et al., 2015, p. 97). Moreover, it is important to account for “values seams—the places where multiple values are held in tension”

(Shilton, 2018, 264), as these are material loci that will impact resulting decisions.

In internet governance, research approaching values and design has pointed to the need to ensure that emerging technologies follow governance principles rooted in decentralized, open, participative and transparent processes (Almeida et al., 2017), and the consideration that privacy governance should consider regulatory issues along with privacy by design and business practices factors (Doneda & Almeida, 2015). Additionally, with a historical lens on the technical development of the internet in its early stages, Sandra Braman has examined technical community privacy discourse in networking design (Braman, 2012b), and elucidating principles discussions in networking internationalization (Braman, 2012a). Laura DeNardis’s works also bring important contributions to the discussion of values and design on the level of protocol

(DeNardis, 2009), and underlying infrastructure (DeNardis, 2012, 2014). In the present dissertation values are key elements to be unveiled in internet interconnection infrastructure considering not only the values embedded at IXPs but also the ones that arise in their absence.

4. Methodology and Methods

The previous section sought to introduce the main areas of study which this dissertation aims to dialogue with and build on. The purpose of this section is to discuss key tenets of science and technology studies, feminist and decolonial scholarship that structure the research 76

methodology or the making of the present research. The following scholarship oriented the selection of the literature above, and the definition of the research questions as well as the methods to conduct the study. The methodology is the lens with which I looked at internet interconnection infrastructure and IXPs.

4.1. Co-production between Society and Technology and ANT

The present dissertation presupposes that technology and society mutually shape and embed each other—a STS perspective that serves as the basis for different formulations and approaches. Sheila Jasanoff has framed this mutual shaping as co-production (Jasanoff, 2004).

Using the categories of scientific knowledge and technology parallel to each other, she states:

“[S]ociety cannot function without knowledge any more than knowledge can exist without appropriate social supports. Scientific knowledge, in particular, is not a transcendent mirror of reality. It both embeds and is embedded in social practices, identities, norms, conventions, discourses, instruments and institutions – in short, in all the building blocks of what we term the social. The same can be said even more forcefully of technology” (Jasanoff, 2004, pp. 2-3).

Significant contributions from this approach are attention to contextualization and how knowledge is situated; consideration of silences and not only explicit discourses; equal concern with stability and instability; and the integration between knowledge and technology production with power (Jasanoff, 2004). In Leah Lievrouw’s words, “[co-production] rematerializes power, so that it is no longer an abstract ‘force’ or institutional ‘structure,’ but is actually instantiated and observable in the physical forms of social practices, relations, and material objects and artifacts” (Lievrouw, 2014, p. 31).

77

Actor Network Theory (ANT) emerges as a fundamental reference to these studies. It is defined by John Law, one of its founders, as “a method for mapping how every object or actor is shaped in its relations” (Law, 2015, p. 10, emphasis in the original). In other words, an actor is understood as a network, and more specifically, “Any element which bends space around itself, makes other elements dependent upon itself and translates their will into a language of its own”

(Callon & Latour, 1981, p. 284). Consequently, for ANT scholarship, power will emerge from the analysis of a web of relations between actors: humans and non-humans, nature and society

(Callon, 1984; Latour, 1991, 2005; Law, 2015). Power is an attribute of a “chain” (Latour, 1991, p. 110), and not of an entity specifically, given that the entities’ identities and characteristics are defined in relation to other actors, also known as actants, that include humans but also technologies, nature, animals, etc.

In this framework, the macro structure of society is not distinct from the micro structure, despite the scale difference; thus, it is not necessary to search for “global causes outside networks” (Latour, 1991, p. 130) to explain their micro dynamics. In fact, ANT has refused the existence of a social structure ruling life. As pointed out by John Law “There is nothing behind the drama shaping it” (Law, 2015, p. 10). This is why, in this approach, “Domination is an effect not a cause” (Latour, 1991, p. 130), which may be unveiled as a certain web of relations is understood.

One of the most influential elements of this approach is given by Michel Callon.

Examining the efforts of some researchers to congregate allies around a strategy to solve the imminent reduction of scallops in a certain region in France, he shows that when scallops do not respond as expected to the researchers’ tactics, the academics lose social support. Fishermen,

78

even contrary to their association representatives, dissent and changed their attitude, fishing all the scallops that were in a built reserve under an experiment. Michel Callon argues that understanding “power relationships means describing the way in which actors are defined, associated and simultaneously obliged to remain faithful to their alliances” (Callon, 1984, p.

284). In his study, researchers, fishermen, scallops are analyzed in symmetry and show constant movement. The roles that “fluctuate” as the situations evolve, and the “unpredictable relationships” among the entities involved elucidate the kinds of formation and discontinuity of social ties which are seen as evidence of power emerging from a “complex web of relations”

(Callon, 1984, p. 201) where humans, institutions and nature are interlocked (Callon, 1984).

Bringing this example to the study of technology, with the concept of “translation” and

“delegation,” ANT authors assume not only a symmetric relation between humans and non- humans, society and technological artifacts, but also understand that there is a continuous interchange between humans’ goals and artifacts’ functions. This happens in such a way that a speed bump, for instance, can be understood as a “delegation” of engineers’ goals in pavement and concrete, and a “translation” of the speed law enforcement into a technique (Latour, 1999).

In other words, the desired action of making drivers slow down is, in this case, not only expressed by a “negotiable” speed limit sign, in which the driver has the opportunity to ignore it.

Instead, the action is provoked by “unnegotiable speed bumps” (Latour, 1999). Thus, by mediating human’s goals, the technique influences human behavior with its own functions and characteristics. As Latour explains with a speed bump example:

The speed bump is ultimately not made of matter; it is full of engineers and chancellors and lawmakers, commingling their wills and their story lines with those of gravel, concrete, paint, and standard calculations. The mediation, the technical translation (…)

79

resides in the blind spot in which society and matter exchange properties. (Latour, 1999, p. 190).

It is the recognition that artifacts assume actions that make them actors or actants, opening a window to investigate their characteristics as such. For Latour, “Responsibility for action must be shared among the various actants” (Latour, 1999, p. 180) because action is conceived not as an exclusive attribute of humans, but as a result of a combination of agents, including technical artifacts. Most importantly, actants are part of a collective; humans and artifacts are embodied in institutions and as so they are an example of a “body corporate”

(Latour, 1999, p. 192), able to act with purpose. According to Latour, “Purposeful action and intentionality may not be properties of objects, but they are not properties of humans either. They are the properties of institutions, of apparatuses, of what Foucault called dispositifs” (Latour,

1999, p. 192).

STS, and ANT scholarship specifically, is a useful research guideline for studying internet interconnection infrastructure as it allows to look at cables, antennas, switches, routers, private and public institutions altogether. According to Latour, “When we talk about a

‘technological infrastructure,’ we are always designating a more or less patched-together mix of arrangements from more or less everywhere that others seek to render irreversible by protecting it from analysis, making it a carefully sealed and concealed black box” (Latour, 2013, pp. 213-

214, emphasis in the original). Nevertheless, the author continues, “everything that has been set up can be broken down” (Latour, 2013, p. 214).

This framework opens the possibility of unveiling values and intentions embedded into technologies to examine how technologies affect society and are mutually affected by it. In shifting the center of the analysis from humans only, power relations become more complex, and

80

their focus can be found at any actor, including IXP design and law, to name a few. Some of the limits of this theoretical framework are discussed below.

4.2. ANT Critics

The principle of symmetry between subjects and objects, and action and intentionality related to non-human actants have been among the main controversies and skepticism with ANT

(Lievrouw, 2014; MacKenzie & Wajcman, 1999). On that, Bruno Latour’s explanations, endeavoring to undo essentialized views on what objects are, are the main arguments to face such critics. For him, objects and people are all “object-institutions” (Latour, 1999, p. 188; 192), from an airplane that ends up being an institution—the airplane company—to a body that ends up being a body corporate: “Real artifacts are always parts of institutions, trembling in their mixed status as mediators, mobilizing faraway lands and people, ready to become people or things…” (Latour, 1999, p. 193). In this sense, the symmetry among humans and non-humans is not focused on their action and intentionality, but it is related with their very being. This becomes clearer with the contributions that other lines of thought bring to ANT, such as feminist as well as postcolonial and decolonial thinking, as will be seen in the next section.

ANT has also been accused of a kind of “Machiavellian relativism” due in part to

“ANT’s stubborn denial to accept anything as given for its political consequences” (Berker,

2011, p. 509). It has also been considered a surveillance mechanism itself for its inclination to

“revea[l] aspects of the phenomenon under investigation that might otherwise go unnoticed”

(Fuller, 2006, p. 7). In general defense, John Law argues that ANT does not assume that 81

“strategies are inevitably explicit or cynical” (Law, 2015, p. 11), and that by understanding how mechanisms of power work, this approach may be used to “undo” them (Law, 2015, p. 11, emphasis in the original).

Possibly the main critique to ANT comes from its relationship with imperialism. Such association is intricate because in the origins of ANT’s “constructivist”23 approach is its opposition to scientific “realism” and universality, strongly associated with imperialism itself.

Realists defend that a scientific fact will be valid everywhere, at any time and place, autonomously. The problem with that, as Rámon Grosfoguel explains, is that “Historically, this has allowed Western man (the gendered term is intentionally used here) to represent his knowledge as the only one capable of achieving a universal consciousness, and to dismiss non-

Western knowledge as particularistic and, thus, unable to achieve universality” (Grosfoguel,

2011, p. 6-7). As a response, ANT and constructivist scholars have challenged the realism universality, defending that reality is not fixed and may be apprehended differently (Fuller, 2006, p. 35). In other words, instead of looking for the “absolute truth,” as realist scholars do, constructivists avoid understanding the world as a “purely objective perception of reality”

(Maxwell, 2013, p. 43).

Whereas ANT and constructivist researchers have countered the realist universalism, the critics they have faced regarding imperialism can be exemplified by arguments emphasizing that

Bruno Latour and constructivists in general are “insensitive to pre-existent (‘structural’ or

‘historical’) power relations between the parties” (Fuller, 2006, p. 39). Steve Fuller contends that

23 Bruno Latour questions the constructivist label received for his work, challenging that “In spite of my reputation as a ‘social constructivist,’ I have always considered myself as one of those who tried to offer another realistic version of science against the absurd requirements of epistemology that could only have one consequence: skepticism” (Latour, 2007, p. 31). 82

by focusing on the current negotiations visible through an object under study, relevant aspects that persist over the years and affect the resulting phenomena can go unnoticed. For him,

“Latour’s actor-network theory canonizes the imperialist’s indifference as constructivist research methodology (…)” (Fuller, 2006, p. 39).

This is the scenario in which critical scholarship, including feminist, postcolonial and decolonial studies, gains space to increasingly influence ANT and science in general. John Law has suggested similarities between feminist scholarship, and current ANT. For him, along with

Michel Foucault’s work, they both “argue that patterns recur: that the world isn’t a different place every morning.” And he continues: “Perhaps (I doubt this is true for early actor-network theory) they are also all saying that there are sustained patterns of inequality” (Law, 2015, p. 11).

Yet, whereas ANT principles and feminist constructivist approach have been convergent in opposing realist universalism, defending plural knowledges and reflecting critically on power dynamics, their overlaps do not confound each other: to be an ANT scholar does not make you a feminist scholar or vice-versa.24

4.3. ANT meets Feminist Epistemologies and “Epistemologies of the South”

24 A prominent feminist scholar’s comment in the 1980s on one of the ANT prominent founders indicates that Bruno Latour was not considered a feminist per se: “Latour's brilliant and maddening aphoristic polemic against all forms of reductionism, makes the essential point for feminists: ‘Méfiez-vous de la pureté; c’est le vitriol de l’ame’ (Beware of purity; it is the vitriol of the soul). Latour is not otherwise a notable feminist theorist, but he might be made into one by readings as perverse as those he makes of the laboratory, that great machine for making significant mistakes faster than anyone else can, and so gaining world-changing power” (Haraway, 1988, p. 596). 83

“Epistemology is any notion or idea, reflected or not, about the conditions of what counts as valid knowledge” (Santos & Meneses, 2009, p. 9, own translation). Such a concept cannot be detached from the scientific realism, or as João Nunes Arriscado explains, from the emergence of modern science and a knowledge theory under which it would be possible to define what is true and false (Arriscado, 2009, p. 217). To oppose this notion, the defense of a multiplicity of epistemologies is necessary to account for the world. Challenging the belief in a prevalent epistemology, then, becomes the very definer of feminist theories as well as knowledge based in non-Western, postcolonial and decolonial theoretical matrices, provoking a fruitful encounter with ANT scholarship. The conception of a “pluriversal” (Grosfoguel, 2011; Sundberg, 2014), instead of universal world, is the common ground for these theoretical frames.

For Bruno Latour, for instance, knowledge acquisition is a “pathway” that allows for entities, when included into it, to gain new “modes of existence” (Latour, 2007, p. 24). That knowledge is not objective beforehand, but only when a community shares that pathway (Latour,

2007). In his words, “Objective knowledge is not first in the minds of scientists who, then, turn to the world and marvel at how their ideas ‘fit’ with the entities out there: objective knowledge is what circulates and then grants the entity seized by the networks [pathways] another mode of existence (…)” (Latour, 2007, p. 30).

Feminist scholarship, even if not a cohesive unit, is in its very nature a discussion of power and dominant narratives that subjugate non-dominant groups. It promotes subaltern perspectives, which are not fixed and cannot be characterized by static attributes—as a male character may be in a subaltern position in a feminist-rooted scenario, for example. To examine power imbalance in society, feminist scholars join perspectives of gender, but also race, colonial

84

heritage and other motivators of violence and exploitation. With that, they add to the study of humans and non-humans solid and historically situated variables that are absent from other theoretical perspectives—SCOT and ANT included.

Sagacious feminist analyses come fundamentally from their problematization of silences.

In this sense, where, for instance, some scholars do not see gender relations, feminist scholars see “hyper masculine” representations, as in the case of colonialism history focused on “heroic navigators, conquistadors, traders, priests, Indigenous chiefs” (Harding, 2017, p. 629), or the approach to technology as a masculine artifact given by Cynthia Cockburn previously discussed.

Feminist theories provide fundamental methodological references to the present work, from attention to silence, to Donna Haraway’s concept of “situated knowledge.” In a thought- provoking piece, she posits situated knowledge as “feminist objectivity” (Haraway, 1988, p.

581), as opposed to the supposed neutral and universal (for her reductionist) objectivity. In doing so, she questions traditional scientific approaches by shifting the attention to who the researcher is and where s/he is positioned to observe phenomena as aspects that matter and cannot be disregarded. In the core, she is concerned with the impact of science on society, our own role in that as researchers, and the respect for different epistemologies.

Perhaps an ethical consequence of such concerns that shapes her constructivist approach is the author’s considerations about the objects of study, or the non-humans as ANT would name them (e.g. living beings, the earth, technologies, etc.). ANT approximates founding feminist conceptualizations in its position against a modern science narrative of the individual-object relationship, where the first is rational and the second is passive. As Haraway points out,

“Situated knowledges require that the object of knowledge be pictured as an actor and agent, not

85

as a screen or a ground or a resource, never finally as slave to the master that closes off the dialectic in his unique agency and his authorship of ‘objective’ knowledge” (Haraway, 1988, p.

592). Taking the previous assertion then to its limit, it is not only the phenomena under study that are influenced by the agency of the objects of study; the scientific production is itself a result of an interrelation—and why not to say co-production—between researchers and their focus of study.

Related to that, Donna Haraway approaches the “vision” as a sense that needs to be reflected on and “reclaimed” (Haraway, 1988, p. 582). Calling attention to the values and the possible power imbalance embedded in the act of seeing, or the “gaze,” she precisely points to the distance that may be established between who sees and who is seen: “The eyes have been used to signify a perverse capacity-honed to perfection in the history of science tied to militarism, capitalism, colonialism, and male supremacy-to distance the knowing subject from everybody and everything in the interests of unfettered power” (Haraway, 1988, p. 581). By critically approaching the seeing, Donna Haraway makes historical patterns an unescapable variable of analysis—an adds to early ANT scholars, as well as other fields of scholarship. Her perspective is in line with perspectives of knowledge production that comes from Latin

American scholars who are similarly critical of the Western narrative of modern science and for whom knowledge production does not suppose hierarchy between a “rational” subject and a

“natural” object. As Anibal Quijano, one of the decoloniality studies founders, puts it,

“Knowledge (…) is an intersubjective relation for the purpose of something, not a relation between an isolated subjectivity, and that something” (Quijano, 2007, p. 173). This perspective

86

can also be found in the basis of ANT’s symmetry assumption and Bruno Latour’s maxim that

“we have never been modern.”

Therefore, there are convergences between feminist scholarship and decolonial thinking.

Specifically regarding decolonial scholarship, Sandra Harding calls attention to the influence of the work of the Chicana and feminist Gloria Anzaldúa (Anzaldúa, 2012) on decolonial thoughts in Latin America (Harding, 2016). There are also influential decolonial and postcolonial feminist scholars such as María Lugones and Gayatri Spivak (Lugones, 2010; Spivak, 1994), whose work has also been considered the respective founding of these schools of thoughts (Bhambra, 2014).

Regarding dialogues between postcolonial and decolonial studies, and despite differences in geographical origins—the Middle-East and South Asia for postcolonial studies, and Latin

America for decoloniality studies—Gurminder Bhambra notes that both challenge Western narratives of the world with “the broader histories of colonialism, empire, and enslavement”

(Bhambra, 2014, p. 115). The term “decolonial,” specifically, has been used to generally refer to the legacy of colonialism in the present time, having as its origin Latin American theorists’ works problematizing the Iberian colonialism in Latin America and hegemonic narratives

(Harding, 2016, p. 1080-1081). I will use the term decolonial in view of its general meaning, but

I consider it useful to bring two crucial explanations from the founding theories on decoloniality that inspire discussions on dependency and the global South as analytical categories that will permeate the present analysis on internet interconnection politics.

Anibal Quijano’s “coloniality of power” concept sheds light on the period after the

Western political colonialism practices, which he emphasizes as being founded in social discriminations based on race, ethnicities, nationalities, and markers in accordance with the

87

historical time and actors associated with them (Quijano, 2007). The author defends that the coloniality of power succeeded colonialism, being a form of continuous domination in that, in the absence of colonial political structures of the past, there is “a colonization of the imagination of the dominated” (Quijano, 2007, p.169). Coloniality of power is thus based on “‘racist’ social relations,” (Quijano, 2007, p. 171) and transcends to economics and the capitalist world system in discussions of dependencies and exploitations.

Indeed, dependency and coloniality have been connected. For Walter Mignolo, another influential voice in decolonial scholarship, “dependency at large was and is the basic strategy in the exercise of coloniality of power” (Mignolo, 2002, p. 62). And even in a world-system where globalization is commonplace, and “center/periphery” categories may be blurred, the author highlights that “the interstate system and the coloniality of power organizing it hierarchically have not vanished yet” (Mignolo, 2002, p. 62), calling attention to the continuous dependency condition among national states or interdependency, if one considers the market vocabulary about transnational corporations (Mignolo, 2002, p. 62, emphasis in the original).25

Yet in this regard, according to Boaventura de Sousa Santos and Maria Paula Meneses in their edited book “Epistemologias do Sul” (Epistemologies from the South, in free translation),

Anibal Quijano contributes to “an epistemic interpretation of the situation of domination present in the global South. For the author, the destruction of the coloniality of power as a relation of

25 Here, Walter Mignolo’s discussion is clearly making reference to the dependency theory originated in Latin America in the 1970s (Cardoso & Faletto, 1979). In Sandra Harding’s analysis on decolonial scholarship, she also mentions other influences in this school coming from Paulo Freire’s pedagogy, economic and political events in Latin America and Indigenous social movements (Harding, 2017). Gurminder Bhambra also mentions influences from the world-systems theory and the Frankfurt School (Bhambra, 2014). 88

exploitation is one of the determining factors in the struggle against the universal pattern of

Eurocentric capitalism” (Santos & Meneses, 2009, p. 14, own translation, emphasis added).

The second important focus of Anibal Quijano’s work on coloniality of power is

Eurocentrism. Quijano shrewdly posits that there would not exist a Western identity and modernity narrative without colonial domination. As Gurminder Bhambra explains, the Western image, or the “European identity,” is a result of “a process of differentiation from other cultures.

Yet, there is little reflection within European social and political thought on how those other cultures constitute the ground of European self-realization (in both senses)” (Bhambra, 2014, p.

118, emphasis in the original). Visibly, the other does not exist actively in Western narratives. In

Quijano’s words, it is “totally absent; or is present, can be present, only in an ‘objectivised’ mode” (Quijano, 2007, p. 172). Nevertheless, the idea of non-Western is constitutive of what

Western is. And the faith in dichotomies, such as human and non-human, and a racialized view of colonizers and colonized—embedded in coloniality (Lugones, 2010)—shape the objectification process, as already approached by Haraway.

The study of internet governance and internet interconnection infrastructure can substantially benefit from the concepts and rationale above. Established explanations of the internet ecosystem centered in Western narratives serve to conceal the role of the South in the current state of the internet, including how infrastructure disadvantages in the South contribute to the emergence of giant IXPs in the North. The concealment of the South has negative consequences in terms of knowledge and policymaking, as it helps keep continuous colonial processes hidden. This is the main reason to include the global North and the global South as symmetric variables of analysis in the present research: situating the Western perspective.

89

4.3.1. Global South

The global South is an increasingly used category in academic scholarship that deserves attention as it has a political meaning and should not be understood as referring to a specific location: “the Global South is contextual” (Wolvers et al., 2015, p. 2), authors warn. The term is a response to previous nomenclatures that implied hierarchical and evolutionary meanings

(Wolvers et al., 2015), such as “Third World” and the developmental discourse, which has implied the need for “developing” countries to take the same trajectories of “developed” ones. In decolonial terms, “Developmentalist discourse offers a colonial recipe on how to become like the

‘West’” (Grosfoguel, 2011, p. 23).

Importantly, for Anne Garland Mahler, global South, as used in cultural studies, raises a

“postnational sense,” by assuming a general condition of marginalization resultant of globalization that surpasses national borders and recognizes that “there are Souths in the geographic North and Norths in the geographic South” (Mahler, 2017). This idea was already developed by Boaventura de Sousa Santos, for whom immigrants, victims of racist and sexist environments, among other underserved groups in the North represent the global South (Santos,

2012). In the same line, according to Alfredo López, the global South concept “marks, even celebrates, the mutual recognition among the world's subalterns of their shared condition at the margins of the brave new neoliberal world of globalization” (López, 2007, p. 3). As used by these scholars, the term embeds a social, political and economic critic into the non-realized promises of capitalism.

For Boaventura, it is necessary to “learn from the South and with the South” (Santos,

1995, p. 508). In the same direction, discussing Postcolonial Studies and STS, Warwick 90

Anderson suggests that Asia should be seen “as method” (Anderson, 2012), or “a site of theory making in STS, not just a space for data extraction or a place to which European concepts diffused. (…) [An] Asia that is good to think with, and think from, in STS, rather than a fixed, hegemonic geographical region or essential civilizational entity” (Anderson, 2017, p. 229).

Altogether, this scholarship calls for a repositioning of the South in the knowledge production to enable new ways of thinking and, consequently, new realities.

In view of both the coloniality phenomenon that characterizes social, political and economic realms of the present, and the need to seeing the Western paradigm as inseparable from its colonialist practices, decolonial scholarship calls attention to socio-historical patterns.

Along with feminist bodies of thought, it requires other sensibilities and additional reflections to be incorporated into the methods and ANT-oriented analysis. The challenge to put these different approaches together is how to recognize socio-historical patterns without compromising ANT capacity to unveil situations by in loco research. The answer to this question lies in research methods and the guidance of inductive techniques that characterize science and technology studies and internet governance scholarship.

4.4. Methods

This research follows primarily qualitative methods founded in an ethnographic approach

(Maxwell, 2013). Following STS empiricist practices that “privilege methods involving face-to- face encounters with subjects in or around the subjects’ native knowledge production sites” 91

(Fuller, 2006, p. 51), the present dissertation is based largely in primary data collected through in-depth interviews and participant observations, complemented with secondary data publicly available regarding internet routing data and IXPs’ worldwide distribution. It is inspired in the possibility of conducting “ethnography of infrastructure,” as Susan L. Star classified her studies on large technical systems usability (Star, 1999; Star & Ruhleder, 2015), yet it extends the focus to critical aspects of internet infrastructure’s deployment, design and governance.

In view of that, the predominant field work approach led the research to follow the people, the infrastructure and the information available, while theoretical literature was a guide to direct the focus, the questions and observations. This is in line with what is kno wn as grounded theory, whose definition is to allow “the discovery of theory from data” (Glaser &

Strauss, 2006, p. 1), and application is recommended to the study of “invisible things” (Star,

2015, p. 125)—as aspects of infrastructure may happen to be sometimes.

Quantitative data is examined as a secondary source of analysis. Together, primary and secondary data helps to substantiate what is governance by internet exchange points (IXPs), approaching research questions that aim to leverage our understanding of current global communication in terms of the rights at stake, infrastructural interdependencies between the global North and the global South, and IXP design exploitation by neoliberal states in line with the emergence of surveillance capitalism.

4.4.1. Primary Data

For conducting ethnographic research, three IXPs were targets: DE-CIX in Frankfurt,

IX.br in São Paulo and IX.mx in Mexico City. DE-CIX is considered the biggest IXP in the 92

world in terms of traffic. It was founded in the beginning of the commercial internet in 1995, and is based in Europe. IX.br in turn was founded in 2004. It is considered the biggest IXP in the world in terms of number of participants and is based in Latin America. IX.mx is a counterpoint.

It is a small IXP founded in 2014 and is also based in Latin America.

In total, more than thirty in-person interviews were conducted following the three sites:

Brazil, Germany and Mexico. Six interviews were also conducted by phone/Skype calls with interviewees based in Mexico, Switzerland and the United States, while one interviewee from

Germany preferred to respond a list of questions by e-mail.

Additionally, participant observations took place in the following public events:

• DE-CIX Summit Frankfurt, Dreieichenhain, October, 2017

• International Forum on Indigenous and Communitarian Media, Oaxaca, August,

2017

• Internet Governance Forum, Guadalajara, December, 2016

• IX Forum, São Paulo, December, 2017

• OECD Telecommunication and Broadcasting Review of Mexico 2017, Mexico

City, September, 2017

At the International Forum on Indigenous and Communitarian Media in Oaxaca, Mexico,

I was introduced to people who were developing internet infrastructure projects in communities in non-urban areas of the country. Following the event, I went to visit a Tseltal pueblo in San

Martín Abasolo, in Chiapas, and two Zapoteco pueblos, one in Guelatao de Juárez and the other in Santa Maria Yaviche, in Oaxaca.

93

Tseltal people are predominant in San Martín Abasolo, while Zapotecos are prevalent in

Guelatao de Juárez and Santa María Yaviche. The population in these territories are part of the

24.4 million people who recognize themselves as indigenous (INEGI, 2016, p. 3), which means more than 20% of the whole country’s population that corresponds to 119.5 million inhabitants

(Flores & Glez, 2015). The localities present different social conditions.

Table 1. Communities’ Demographic Information

State Location/Town Population Social disadvantage level Internet (2005)a (2012)b Infrastructure Chiapas San Martín Abasolo 2.884 High Community initiative Oaxaca Guelatao de Juárez 544 Very low Municipality initiative Oaxaca Santa María Yaviche 611 High Company initiative a Data organized by the Secretaría de Desarrollo Social (SEDESOL, 2005a, 2005b, 2005c). b Index created by the Consejo Nacional de Población (CONAPO) to measure “a set of social disadvantages of a community or locality” (Vega Estrada, Téllez Vázquez, & López Ramírez, 2012, p. 11).

Demographically, these territories, with less than 700 inhabitants, are not atypical in

Mexico. According to data from Instituto Nacional de Estadística y Geografia (INEGI)

(National Institute of Statistics and Geography) organized in 2015, approximately 190 thousand localities have a population fewer than 5,000, which corresponds to 32,4 million or 28,8% of the whole Mexican population (Flores & Glez, 2015).

Regarding access to the internet, both states, Chiapas and Oaxaca, are considered areas with the lowest connectivity rates in Mexico. According to estimations available, between 0-20 out of 100 inhabitants have access to the internet at home. As a comparison, in Mexico City, this number corresponds to 76 out of 100, and in the country, to 43 out of 100 (IFT, 2017, p. 29).

A focus on infrastructure deployment, design and governance oriented the in loco research. Nevertheless, while exchanging experiences with people, I would hear expectations 94

that the knowledge generated during the research would bring social benefits to the communities in the future. I was also informed about the need to expand my focus as a researcher, given that their lives, concerns and interests as communities were much more than the internet projects that

I was looking at. Dennis Jair Mendonza would take me to visit local museums and theaters, and to attend music classes at a public school. He would also open his house and introduce me to his mother—proud for his son to be the first to go to college at the most well-known university of the country. Osvaldo Martínez would take me to traditional parties and would ask me to take care of his daughter and niece while he was feeding some animals in the field. I would go with the children to school activities, and would visit some neighbors with Joaquin Yescas Martínez.

Along with Mariano Gómez, I would be asked to wear traditional clothing during a network set- up activity and be a reason for laughter because of that, and would have my attention caught in lots of good and meaningful conversations that would make the time slow down and make life more synchronous and respectful with the sunrise and sunset.

The approach and time staying in each community varied according to the availability of my hosts, who not only guided me through their communities and infrastructure sites, but also received me in their houses, from one to three days. The visits and fieldwork research at the communities took place in August 2017.

In Mexico City, Frankfurt, Salvador26 and São Paulo, I also visited data centers hosting internet exchange points’ infrastructure, and accessed the area where the core IXP technologies

26 Salvador is the capital city of Bahia, a state in the Brazil’s Northeast region, where a pioneering project named OpenCDN, to be explained in Chapter 6, was in development at the time of visit. 95

are co-located when allowed. Engineers responsible for the IXPs guided me through these visits:

Ibirisol Fontes Ferreira, Italo Valcy, Jorge Ramírez Angón, and Thiago Bomfim.

4.4.2. Secondary Data

Information about IXPs is scattered. There are some public databases built with different collection methods and that have, consequently, differing data available. In order to have a consolidated dataset, between February 2017 and June 2017, I worked with the sociologist and statistician Daniel Ribeiro to merge four popular databases in the field: PeeringDB, Euro-IX,

Packet Clearing House (PCH) and Telegeography. The procedures followed these steps: first, we selected the two largest databases to apply a formula with the goal to match the IXPs records.

Records that did not match were manually analyzed according to name, country, city, website and other variables available, as the wording could differ from one dataset to another. Then, the same procedures were conducted between the resultant dataset and the third and fourth files. The final dataset will be explored in Chapter 4.

The collection of internet routing data, in turn, responds to the goal of looking at big IXPs from the global North and South symmetrically, through the data that circulates in their infrastructure. It is a code ethnography, in the sense that I am not native to the world of BGP data, but want to understand it and organize it according to social sciences categories. For that,

Tiago Gonçalves processed the data, and along with other BGP and network engineers, guided me through it.

A final consideration is necessary to explain the global North and global South categorization. As the global South is a political concept that unites conditions of marginalization 96

and negative results of globalization beyond national borders, in practical terms, there is not a list of countries defined as the global South (Oppermann, 2018). And although international organizations’ publications use the term, their classifications continually refer to “developing, transitional and developed” countries (United Nations, 2018; United Nations Conference on

Trade and Development, n.d.), showing the difficulty to operationalize the term. As used in academia, the global South is a prescriptive concept.

Considering that 1) the academic references for the use of the term point to it as a metaphor to face the consequences of colonialism and capitalism (Santos, 2012); 2) such criticisms are built on the recognition of the economic differences among regions in a global scenario; and 3) the purpose of the present dissertation of using the global North and global

South as variables not only in qualitative, but also in quantitative analyses, and thus as methodological foundations requires a list of countries, this research uses the annual classification of economies released by the World Bank in June 2018 (World Bank, 2018) and categorizes it in a global North and global South framework. The World Bank classification has four categories: High Income, Upper Middle Income, Lower Middle Income and Low Income, and is based on the Gross National Income (GNI) informed by the countries. Below the difference between High Income economies and the other categories is clearly shown. These economic disparities support the research choice of categorizing High Income economies as global North and the other bottom categories as global South.

97

Figure 2. Gross National Income (GNI) by High, Upper Middle, Lower Middle and Low Income Economies. Source: World Bank27

Additionally, in order to have the most comprehensive list of countries possible, I proceeded a comparison between the World Bank income-based classification with 219 economies, and the United Nations 249 countries or recognized areas. Merging both sources, the list of countries used as a reference for the present research ended up with 253 countries plus

European Union. United Nations countries added to the final dataset were defined as global

North if classified as developed, and global South, if classified as developing.

Particular methodological details will be given in the chapters as needed.

27 Available at: https://data.worldbank.org/indicator/NY.GNP.PCAP.CD?locations=XM-XD-XT-XN 98

CHAPTER 2

INFRASTRUCTURING TOWARDS DIGITAL CITIZENSHIP: THE PATHS OF LOCAL INDIGENOUS NETWORKS TO THE GLOBAL INTERNET

In an infrastructure you are never alone: no node is an island.

—Susan L. Star and Geoffrey C. Bowker, How to Infrastructure (p. 235)

This chapter frames the internet as a global infrastructure that is subject to continuous development locally; it explores the emergence of shared networks in indigenous communities as a result of the interaction of humans and non-humans; it frames local people involved in infrastructuring as internet codesigners; it discusses epistemologies from the locality and elucidates the tensions between global and local entities when interconnection infrastructure is taken into consideration; it adds to the understanding of infrastructure visibilities, which occur when infrastructure is paradoxically absent.

1. Introduction

In deciding to begin this dissertation with the internet connectivity experiences of Tseltal and Zapoteco communities, the invitation is to think with them, to understand the global internet and its interconnection ecosystem from both the strategies that these communities have designed and the actors that support their actions to “infrastructuring” the internet.

I borrow the term “infrastructuring” from participatory design scholarship, where

“information infrastructure is viewed as constantly ‘becoming’” (Karasti, 2014, p. 3), in that design is a continuous activity, a “process of inscribing knowledge and activities in new material 99

forms” (Karasti, 2014, p. 3, emphasis added). Such understanding is in line with Paul Dourish’s

“call for an examination of the relationship between infrastructure and experience,” in which the author wants to emphasize “attention to the processes by which digital experiences are produced; and, further, to warrant an investigation of the practices of technological design that generate these arrangements” (Dourish, 2015, p. 185, emphasis in the original). The concept of infrastructuring captures such a focus, and even beyond these parameters. By considering design as a process, internet infrastructure design may be investigated encompassing practices where non-specialized people are involved. In doing so, it complements perspectives that have focused primarily on network engineers’ practices (Abbate, 1999; Braman, 2012a, 2012b).

The analytical benefits of examining internet infrastructure design as a process are threefold. First, it sheds light on highly specialized people (e.g. companies’ engineers) and non- specialized people (e.g. individuals with any background) as internet codesigners. Second, it contributes to visualizing more clearly an important dimension of infrastructure that refers to its path dependencies, as it continuously emerges from an already established underpinning, and

“inherits strengths and limitations from that base” (Star & Ruhleder, 2015, p. 381). Finally, by highlighting the tangible efforts to enable the internet locally, such approach counterpoises the ethereal aspects that populate the imagery of a global and borderless internet. This is one of the moments when the local becomes global and the global, as an assemblage that connects many locals, comes to exist.

Understanding local access sharing as part of internet interconnection is a consequence of these facts. When discussing IXPs and the networks connected to them, processes of infrastructuring happening in the ground—and the multiplicity of actors involved at that level–

100

become invisible. As we shall see, I want to make the case that interconnection occurs not only at IXPs, or privately at corporations’ locations. It also happens at arrangements closer to end users under unbalanced relations of power between established and emerging ISPs. Although still overlooked, these interconnection points become part of the internet infrastructuring and have an important role in the midst of local and global internet infrastructure.

2. Moving Through a Visible Route from Oaxaca to Santa María Yaviche

Arriving in Santa María Yaviche from Oaxaca was a deep journey in itself. With only one bus available per day, seats become scarce resources, especially on religious feast days, as was the case that Monday. A good option then was to arrange new seats in the back of the bus using luggage, pail and pillow. After a gentleman casually reading a paper enroute to visit his family got off, his pail was quickly occupied by a lady who entered the bus with a friend and her child. Mother and child, on their turn, found a seat on the ground on my side, and a neck pillow served her as a back pillow, facilitating carrying the two years old boy. Both women were coming back from the Ixtlán market where with some regularity they sell bananas, beans and other products.

Although the distance that separates Oaxaca, the city capital, from Yaviche is not more than 129 km according to online maps, the route, with mountains, curves and numerous stops, takes more than eight hours. The distance, therefore, is best defined by time instead of by physical distance. In the bus, many people were standing in the aisle, while many others were waiting at different points of the bus route, making it difficult to count how many passengers entered the bus each stop and how many more could fit in it. 101

Figure 3 and Figure 4. The transportation system from Oaxaca to Santa María Yaviche28

Conversations nonetheless make the long trip lighter. Fortunately, many people around were bilingual; while among them most communicated in Zapoteco, they spoke with me in

Spanish, disregarding my non-native speaker accent. I was asked many questions about my life, my family and what I was going to do in Yaviche. I explained my interest in learning about internet connectivity in the region to facilitate communication among people. This explanation was responded to in indecipherable silence by that mother. I filled it in by saying that the transportation system was also important and should be improved, which was responded to very quickly: a resounding “yes!” For Luisa,29 with whom I spent most of that time, work opportunities and transportation were more urgent issues. That was the second trip that she had taken sitting on the ground that day. She is a non-internet user, and with difficulties to work in her community, she had come to sell her products in the morning with her child.

In the same bus, late at night, my conversation with an elementary school teacher showed another perspective. Geraldina became very excited about any possibility of improving

28 These and the following photographs were taken by the author. 29 This and the other passengers’ names are pseudonyms. 102

communication in the region. She commonly spends more than 5 dollars (96 pesos) per month to have access to the internet on her cellphone two times per week. But again, this was in contrast to the reality of a very well-informed elderly person seated close to her who was a non-internet user. Rubén was proud of “all the projects” that my host in the community, known as the

“engineer,” was conducting, and about which he watched a video on the internet during a visit to his daughter’s home in the capital Oaxaca.

Although it is not possible to estimate how many different profiles of internet users and non-users could have been disclosed during that trip, this experience reveals a spectrum of significance of internet connectivity to people’s lives in the area. With the common absence of the state policy in many areas, it may go unnoticed that the Mexican Constitution declares “The

State shall guarantee access to information and communication technology, access to the services of radio broadcast, telecommunications and broadband Internet” (Mexico,

1917[2015], Art. 6).

Moreover, what was in context in a transportation system, where people travel from one point to another, serves as a metaphor for an information circulation system, where people, converted into voice and text, are encapsulated into data packets and traverse many points until they arrive at their destination. Scarcity of the transportation/broadband services and the limited speed when they are available could also be added. And although such analogies have numerous limitations, particularly that the decisions about where people/packets go next are taken differently—through a predefined route in the first case, and a defined-on-the-way-route in the second—the comparison is intended to call attention to the materiality of internet infrastructure in Yaviche. To arrive in town, in the house of my host in the community, such infrastructure

103

similarly follows a route towards Oaxaca, where interconnection with the larger internet will occur.

3. Case 1: Internet Infrastructuring by a Regional Commercial ISP

Joaquín Yescas Martínez, 17 years old, was a speaker at the International Forum on

Indigenous and Communitarian Media in Oaxaca, where I met him presenting information about

Santa María Yaviche’s community cellular network. At that time, he was a high school student, preparing to be an undergraduate in the community’s brand-new university, CEUXHIDZA

(Centro Universitario del Pueblo Xhidza). With the goal of earning a bachelor’s degree in comunalidad (communality), he aspired to build a computer operating system (OS) in

Zapoteco—the language spoken in Yaviche, and the idiom of 6.5% of the indigenous population who speak a native language in Mexico (INEGI, 2016). Joaquín was also an announcer on the communitarian radio, Aire Zapoteco Bue Xhidza (Zapoteco Air) and was in training to be the local technician for the community cellphone network started and coordinated by his uncle,

Osvaldo Martínez, the agronomist engineer of the community.

Joaquín and Osvaldo hosted me in Yaviche, one of the more than 20 communities that form the Telecomunicaciones Indígenas Comunitárias (TIC A.C.) (Indigenous Communitarian

Telecommunications). The organization remarkably acquired the license of the frequency band of 847-849/892-894 MHz (Mega Hertz) from the national telecommunications regulator, the

Instituto Federal de Telecomunicaciones (IFT) (Federal Institute of Telecommunications), for indigenous social use (IFT, 2016). For that, they counted on the legal support of the NGO Redes por la Diversidad, Equidad y Sustentabilidad A.C. (Redes) (Network for Diversity, Equity and 104

Sustainability), and the technical support of Rhizomatica, organizations well known for their achievements in Mexico.

With the social indigenous license, the first of its kind state regulation worldwide (Belli,

2018), and local investments of approximately 5,000 dollars—in towers, antennas, telecommunication equipment such as BTS (Base Transceiver Station) and BSC (Base Station

Controller) 30—along with a contract with a regional Internet Service Provider, Yaviche municipality started its own community cellphone network, independent of traditional telecommunications companies for local communication. The organizations involved call this kind of initiative a “community network” because from the start, the project is approved in the community assemblies, the costs to maintain the infrastructure are shared by the users, and the administration of the network is conducted by community’s representatives.

At Yaviche, at the time of this writing, approximately 40 users are paying less than 2 dollars (32 pesos) per month to have unlimited calls within the network, and approximately less than 2.5 dollars (40 pesos) per minute to call to other cellphone networks nationally and internationally. These differences in price help explain the design of this innovative cellphone network. It has its own equipment rather than just leasing a line from a local telecom, which allow unlimited local calls for a relatively low cost. Unlike that, to connect to cellphones from other networks, internet is needed and it is charged per minute.

Regarding sustainability, due to the sharing of costs, a community cellphone network requires a minimum number of users to enable financial network maintenance. Below the

30 For technical-oriented guidelines on community cellphone networks fomented by Rhizomatica and Redes A.C., see Huerta (2016). For broader perspectives from these organizations’ founders, see Bloom (2015) and Huerta (2018). 105

threshold of 80 users, the financial situation in Santa María Yaviche has become more difficult to sustain, in part because of the basic internet cost of approximately 88 dollars (1.500 pesos) per month.31 The service provides 8 Mbps (megabits per second) of download and 2 Mbps of upload.

As a comparison, in Mexico City, 10 Mbps of download costs approximately 20 dollars (349 pesos) per month, more than four times cheaper for a higher speed. The interconnection of the local cellphone network to the larger telephone system is built, by design, on costly internet services. Thus, for community cellphone users to talk with acquaintances, friends and family served by other cellphone networks elsewhere they depend on the internet arrangements available, including the regional ISP’s market strategies and interconnection arrangements to the global internet.

In 2017, Protokol Telecomunicaciones was the only ISP in Santa María Yaviche, attending to the community cellphone network under a commercial agreement in which inter- network calls via Voice over Internet Protocol (VoIP) are made possible. Protokol started in the market in Oaxaca in 2008, selling internet mainly to cyber cafés and communitarian computer centers. Since then, their business has evolved to provide internet services to families at home; however, the company’s main income nowadays is derived from hourly charges to individuals who want to use the internet in mobile devices through Wi-Fi hotspots. Protokol has focused on providing services in rural areas, mainly the territories of Mixes and Zapotecos, reaching more than 90 of the 570 municipalities recognized by the federal government in the state of Oaxaca.

Interestingly, as a regional ISP, it works in a niche not served by big telecommunications

31 Other important costs are the electricity bill, which is approximately 47 dollars (800 pesos) per month, and the contribution to TIC A.C., 15% of the income. 106

corporations for being considered not profitable. Remarkably, while Oaxaca has 23% of the municipalities of the country, it corresponds to 3% of the population in Mexico (Bloom, 2015, p.

19).

Most of the communities in the state of Oaxaca follow the political system of “uses and customs” whereby, among other characteristics, decisions are made in assemblies and the territory is of common use, setting barriers for private companies to buy or rent land. While the regional political characteristics have been considered fundamental for deploying community telecommunication projects (Bloom, 2015), it requires compliance by the private sector with the rules of these territories and negotiations. For instance, Protokol’s founder, Alvaro Arroyo, has created what he calls “social exchange agreements” with the communities, through which, in total, approximately 15% of the companies’ revenue is shared with them in the form of free services for public buildings and some households. Prices with communities can also be subject to negotiation.

At the time of this writing, the company has approximately 1,700 fixed broadband clients, a threshold difficult to increase, Alvaro Arroyo says, given the current prices and demand characteristics. In response to this financial threshold, the company has invested in mobile internet:

We saw that the only way to grow [our business] in la Sierra [Juárez] was catching people who did not have computer equipment in their house, only cellphones or tablets.

Since then, Protokol has invested in installing hotspots in the communities, on a pay-per- use basis.

107

3.1. Regional ISP’s Local Infrastructure

Protokol arrived in Santa María Yaviche building a wireless network of towers and radio antennas. It sells to their clients the customer-premise equipment (CPE), which is also a radio antenna commonly referred to as “radio” only. The radio captures the Wi-Fi signal in the households’ rooftops and is connected to a router at the home. The cost of such devices, including the installation, is approximately 118 dollars (2.000 pesos), added to the monthly bill, making internet sharing and reselling common strategies among end users.

The electromagnetic spectrum, whose management is regulated by the national government, is a crucial part of the communication infrastructure. This happens because frequency fractions of the spectrum serve as a medium for the transmission of radio, TV, mobile and Wi-Fi signals (Foditsch, 2017; ITU, n.d.). As spectrum management affects the course of media development, civil society has been increasingly active in discussions of spectrum distribution (Foditsch, 2017), including of unlicensed spectrum as the Wi-Fi, which is located between radio waves and microwaves “used to listen to the game, and cook your dinner, respectively” (Neal, 2013). Players other than big telecommunications and TV corporations have also been involved.

Protokol acquired the regulator authorization to use an unlicensed frequency for commercial purposes. With that, the company can offer its broadband services without having to pay for using the spectrum to the national government, a way to eliminate a barrier to arrive as communities such as Yaviche, where big companies, which have spectrum allocated, say that is not profitable to reach. Alvaro Arroyo notes how spectrum is a crucial technological resource for his business: 108

I am a hybrid between communitarian and commercial (…) To transport through radio [frequency] has technological limits and high costs. The spectrum is the most expensive one.

Beyond regulatory and economic challenges, the affordances of an invisible artifact as the spectrum become salient when the quality of the internet service is affected by interferences due to saturation of a frequency band. Companies report up to 40-50% of throughput lost in the transportation network, which means that people in a rural community can purchase a link of 10

Mbps but may end up having access to only half of it. In Arroyo’s words:

Interferences cause data packets to be transmitted more than once until the devices receive them correctly. Then the throughput that would be 100% to a fixed band rate decreases, for example, to 40%. That means that if the normal bandwidth of the channel that you use for communication of 20 MHz has a maximum of 100 Mbps of speed, it decreases 40 Mbps.

Some factors for the loss of transmission speed and quality are 1) noise in the transmission due to many other parallel communication transmissions on that frequency; 2) the low quality of equipment used by networks at that frequency; and; and 3) the positioning of the devices, meaning how clear is the line of sight among them. The characteristics of the spectrum, the Wi-Fi propagation and the other technological resources used to build local networks together constrain people’s actions and accessibility in this scenario.

The most popular unlicensed frequency, commonly used in mobile communication, is the

2.4 GHz, and increasingly the 5 GHz. Protokol uses the latter for distributing its internet services, but the growing use of both, including by emerging community and commercial small players, has led companies such as Protokol and social organizations such as Redes to advocate for the availability of another frequency band for improving internet access in indigenous communities, particularly that the regulator should issue a frequency license that can be administered by the communities—similarly to what was already done with the 847-849/892-894 109

MHz band licensed for the purpose of building community cellphone networks. Such a frequency would be used for strengthening small ISPs, reducing their costs and improving the quality of their services. According to Erick Huerta from Redes:

We would not use this spectrum commercially. We would give access to it freely, as if it was of free use, but under a scheme of collective governance. That would be a social indigenous license.

By “collective governance,” the specialist means something similar to what occurs nowadays with the community cellphone networks, in which any decision needs to have been taken by the community assemblies. This type of policy would certainly benefit not only ISPs as

Protokol, but also those which currently work without a license because of the difficulties and costs perceived to have one. These non-licensed businesses similarly require their clients to purchase the radio antennas that will be installed in their rooftop as well as the router to be installed in their homes, and do not put restrictions to share the internet signal once they contract for it. Representatives of an ISP in this situation recognize that the end users would share the signal even without permission, since nowadays it is very easy to find and buy a device to do so.

Including non-licensed ISPs to this scenario is important because although they have similarities with a formal ISP, as they build a wireless infrastructure with towers and radio antennas to reach their clients, these providers show how their legal status and economic constraints, lead to a different interconnection arrangement between its network and the internet when compared to Protokol.

3.2. Regional ISP’s Interconnection with the Larger Internet

110

Protokol connects to the larger internet by buying the service from a larger ISP known as

TotalPlay. Protokol purchases a dedicated link of 1 Gbps (gigabit per second), which ensures the same speed of download and upload. In this commercial relationship, TotalPlay assumes the role of distributor of carriers—a role that arises from the context, as it is just a bigger ISP than

Protokol. Protokol in turn provides what is called “last mile” among telecommunication companies, the final network patch that reaches the end user. Conversely, Erick Huerta, from

Redes, prefers to call it “first mile,” as a way to calling attention to where communities are situated in the internet ecosystem.

As a licensed ISP approved by the telecommunications regulator, Protokol could in theory interconnect to the larger internet by purchasing directly from traditional distributors of carriers, such as Telmex, the Mexican telecommunications incumbent, and Enlace TP—both carriers physically present in Oaxaca. The difficulties for that, however, according to the company’s founder, is that the closest node available for a possible Protokol interconnection is approximately 50 Km away, and the investment in optical fiber cables to reach that node is not economically viable. Another impediment is the price charged by the telco incumbent to sell internet connection to a small ISP. Telmex is Protokol’s potential competitor, given that it works as an ISP as well, even if Protokol is working in niches reneged by big ISPs.

This kind of incumbent’s commercial behavior is not completely unexpected. As recognized in one of the recommendations of the telecommunications regulator’s Consulting

Board—in the context of a government project of mobile network known as Red Compartida

Mayorista, “There are clear incentives for an agent located in a link in a productive chain to seek

111

to displace its customers or users when it also participates in subsequent parts of the chain” (IFT

- Consejo Consultivo, 2015, p. 8, own translation).

The relation between Protokol, a regional ISP or a Tier-3 network, and a national ISP as

Telmex, a Tier-2 network, illustrates the hierarchical organization between businesses that own different portions of infrastructure. As an incumbent, Telmex was responsible in 2016 for 57.7% of the internet provision in the country (IFT, 2017, p. 27), together with Telnor, both sister companies under América Móvil’s control. Its market power to define commercial relations is an important factor of interconnection politics, which will be further explored in the next chapter.

With a small number of players, and the consequent inexistence of a public point of interconnection built on the understanding of mutual benefits—as an IXP is commonly started— the space for generating more affordable connectivity options in small rural communities is highly constrained.

Unlicensed ISPs arise in this scenario, positioning themselves as unofficial and non- declared resellers of big telcos. They build their own infrastructure and attract customers with their own companies’ names. In this way, these emerging ISPs become internet networks, but as they do not have licenses and autonomous system numbers, they are invisible when the internet ecosystem is observed from above. Along with indigenous networks built by municipalities and social organizations, they have specific interconnection models that sustain their invisibilities: they act as residential clients of larger ISPs at ordinary houses in regions served by internet services, and deploy their own wireless infrastructure from those houses. In such a context, households are transformed into access points to the larger internet, under unbalanced relations of power as will be examined in the following sections.

112

4. Case 2: Internet Infrastructuring by a Municipality

Guelatao de Juárez is known as the land of Benito Juárez, a politician with indigenous origins who became Mexico’s president in the nineteenth century (1857-72) (Mexico, 2018). The municipality, of less than 600 people, has Zapoteco roots and is recognized for its higher levels of development in the region, offering a brand-new cinema, a museum, a municipal music band and, since 2014, internet service administered by the local authority.

Humberto Morales Sancris is a network technician at the Universidad Sierra Juárez, and the municipality coordinator of the internet project deployment since 2015, when he was the

Treasurer of the municipality. Municipal jobs are non-paid and part of the required community work in these territories, also known as tequio (Mendonza Bautista, 2017). Humberto Morales explains that before developing their own wireless network to bring internet service to the community from another city, they tried at least two other alternatives. First, they requested, unsuccessfully, service improvements from Telmex, the only ISP available in town until 2014.

The internet service offered was slow and unsatisfactory, delivering actual 100 Kbps (kilobits per second) of download and 10 Kbps of upload. According to him, in response to the request, the company indicated that there was no financial incentive to improve their service. Alternatively, the community representatives sought to negotiate the use, also unsuccessfully, of Comisión

Federal de Electricidad (CFE) (Federal Electricity Commission) infrastructure, which has a substation close to the municipality. CFE deployed its own optical fiber network, which has been used in administrative support of its electrical system, and to connect CFE’s commercial and client offices (Flores-Roux, Mariscal, & Aldama, 2009).

113

With these requests unfulfilled, Humberto Morales, together with Saúl Hernández

Marcial, Héctor Juan Miguel, and Julio García Márquez at the University of Sierra Juárez, decided to deploy a public wireless network, under the municipality responsibility, that would allow them to contract a better service in the capital Oaxaca, and bring the signal to Guelatao. As the Municipal President, Juan José García Ortiz explained:

Given the poor service provided by Telmex, until recently the only internet provider in the region, the city council approved to build an infrastructure that allowed to contract Internet with another provider, establishing a community company capable of providing high speed internet to the [municipality] inhabitants at costs below half the price that is handled in the market. (Bautista, 2017, p. 66)

The initial endeavor of the technical team was to study cases and technologies online, focusing on brands’ websites that focused on “do it yourself” projects. An example is the website of antenna manufacturer Ubiquiti, where one can locate some points of interest on a world map and calculate the signal reach. The price of the radio antennas available is also attractive; there are brands which are offered for approximately 300 dollars (5.000 pesos). Morales notes, however, that despite the grassroots appearance of the project, a basic knowledge of voltage, servers, switches and network is necessary.

The relationship between the affordances of these technologies and physical geography is important when bringing the internet to underserved locations. Humberto Morales and his colleagues soon discovered that this is a task inherently based on trial and error. For example, high geographic points for radio antennas are beneficial for signal capacity, but also are highly vulnerable to electric discharge, damaging the equipment.

After that incident happens, the technicians conducted new tests to identify the optimum location to install the two antennas, one to capture the signal from where it is already available, the capital Oaxaca, and the other to emit the signal to Guelatao. They determined that Yuvila, 114

part of Ixtlán de Juárez, is a kind of middle point between Oaxaca, from where the internet signal was being bought, and Guelatao. Considering a straight line, the link from Oaxaca to Yuvila travels 17.54 Km, and from Yuvila to Guelatao, 19.80 Km. Besides the towers and the cables, four main antennas are necessary for the ecosystem to work, one in Oaxaca to send the signal, two in Yuvila—one facing Oaxaca to receive it, and the other one facing Guelatao to resend it— and finally one in Guelatao, in a tower from where the signal is received and is distributed to reach households and public spaces such as the school, the health center, the movie theater, etc.

Sectorial antennas are jointly used to allow the signal sharing.

As a result of this project, network access improved from the nominal speed of 1 Mbps of download promised by Telmex in Guelatao, only 100 Kbps as measured, to 60 Mbps. The devices and antennas work with the frequency of 5 MHz, in order to be more stable and have fewer interferences—a problem reported not only by companies in the region, but also by named community networks in other parts of the world (De Filippi & Tréguer, 2016). Humberto

Morales has the concern that if they use the frequency of 2.4 MHz, more devices, cellphones and laptops, would be able to detect their network and make it more vulnerable in terms of security.

115

The improvement costs were worth it. The municipality charges approximately 9 dollars

(150 pesos) per household for connectivity. The monthly cost for the infrastructure built from

Oaxaca to Guelatao includes approximately 76 dollars (1.300 pesos) paid to izzi, the Internet

Service Provider contracted in the capital; 12 dollars (200 pesos) paid for the electricity and the rent of the rooftop on a house in Oaxaca—further explained in the next section; 88 dollars (1.500 pesos) paid for the municipality that owns the land in Yuvila. The total is 176 dollars (3.000 pesos). As of 2017, there were 30 households connected, in addition to the public primary school, the public health center, the public library and the city hall, where free internet is offered:

1.5 Mbps divided between the users.

Figures 5 and 6: A household with an external radio antenna and a router inside.

116

4.1. Municipality Network’s Interconnection to the Larger Internet

The first step for a municipality as Guelatao de Juárez to enable internet in their community is to identify the nearest Internet Service Provider already offering internet service.

Community representatives then contract with that service as if they were a residential client. To do so, they commonly make agreements with acquaintances or friends who have houses where the service can be requested and installed. As explained above, they then buy the necessary equipment and initiate collaborations with a party or parties who have the passive infrastructure—towers, posts, etc.—to install the devices in the highest and closest points possible to their own municipalities, ensuring optimal signal capacity. Such collaborations can involve local households that are well positioned, or companies, which sometimes have towers already installed for different purposes, such as television service. The agreements can involve money, but can also be an exchange of resources that benefits both parties.

Interestingly, this model of enabling internet where interconnection means to contract the service of an established ISP is also the strategy that emerging ISPs, committed to enable internet in rural communities but devoid of license, use. They provide internet services obtained from bigger ISPs by sharing the service contracted as if they were a residential client. For that sharing, they engage in infrastructuring wireless networks to reach the communities, similar to what the

Guelatao municipality shows. Different from licensed regional ISPs (e.g. Protokol), these emerging businesses have a smaller capacity for investment and cannot afford buying internet access from a carrier distributor in a business-to-business arrangement. Instead of being a business client of a Tier-2 network, they are a residential client.

117

These experiences are based in sharing by design, as they share a signal initially contracted for one household use and may also allow that clients share the contracted service.

This is also what happens in the subsequent case, led by a social organization.

5. Case 3: Internet Infrastructuring by a Social Organization

In communities where neither cellphone nor internet service is available, the impact of infrastructuring networks autonomously has generated an expanse of recognition, from neighbors to international organizations. In San Martín Abasolo, in Chiapas, Mariano Gómez, an elementary teacher of indigenous education, is one of the Colectivo Ik'taK'op’s (Collective) founders who has put enormous effort into building a wireless internet infrastructure from a town where internet is already available in its own community. In 2017, Mariano was one of the twenty-five awardees of Internet Society, young people below 25 years old recognized by their work for the internet. With him, at least 4 more people are part of the Collective: Néyder Darín

Dóminguez Hernández, middle-school teacher, Luis Ramon Alvarado Pascacio and Osmar

Alfonzo high-school teachers, and Raul Gómez Méndez, who works in the field. Mariano’s family as well as other individuals in the community also support the Collective’s actions.

A visitor to his home will first see a computer lab in the front room, or a “cyber,” as it is known in the region, which has been usually empty. That could be considered a sign of lack of demand, but conversely, it reflects the Wi-Fi services that the collective offers based on the design of the network built in the community. The ubiquity of the “Wifi Zone” that they announce on the wall explains people constantly arriving at the cyber with Pesos Mexicanos coins and leaving with a piece of paper, without using any of the computers there. The paper 118

contains a password that allows one to connect to the Wi-Fi in some parts of the community, where antennas in households’ rooftops work also as hotspots. This is one of the project’s strengths in Abasolo, which can offer internet access for a small amount and connect 20 to 60 devices through these Wi-Fi hotspots around the community.

In Abasolo, the project started buying 10 Mbps of download from a Tier-2 ISP that serves

Oxchuc, some kilometers away from the community as will be further explained below. There are plans to upgrade it to 20 Mbps as of this writing. Among other reasons, one of the motivations for that improvement is the fact that the signal loses quality in transmission, decreasing from 10 Mbps to 7 Mbps when arriving in the community. As in Santa María

Yaviche and Guelatao de Juárez, the internet infrastructure of San Martín Abasolo is built on the unlicensed frequency of 5 GHz, used also by other organizations in the region, causing occasional interference. This fact, compounded with the characteristics of equipment and installation along the network, determines the internet quality that arrives in the community.

Additionally, because by design each device installed is also a hotspot, the more neighbors that have home internet access, the more people can have access to the internet outdoors with the Wi-Fi password offered not only by the Mariano’s cyber café, but also by any person who requested the installation of the antenna at home. In other words, the network design in Abasolo enables a new node to also become a provider, charging for it if the neighbor wants.

Transmission from nodes can reportedly reach approximately 500 meters, a distance that can be affected by the devices used. People from nearby communities have also benefitted from connectivity, once technical analyses show viability for establishing antennas there. The arrangements can vary, but commonly interested people provide and install the towers, while the

119

collective provides the radio antennas. The towers are commonly made of bamboos and provided by the home’s owner that will receive the antennas. Below, the figures show the first time that the internet was enabled in a more distant community known as San Martín.

Figure 7, Figure 8, Figure 9 and Figure 10. Bringing internet to the community of San Martín for the first time.

120

The Collective is engaged in creating an inter-net, connecting some communities around

Abasolo to their network, extending the coverage of the Wi-Fi and the possibility of mobility for people from the community. They are aware that the internet can lose quality with the increase in users; however, having more people connected is a value of the project, so parallel to that, they rely on the low broadband consumption in the community. Mariano Gómez notes the difference in broadband consumption between technology for “communication,” including messages and online calls, versus that for “entertainment,” including video streaming. Restricting the bandwidth that people can use justified by the notion of scarcity and the idea of collaboration emerges as a policy to support connectivity expansion based on sharing mechanisms.

The monthly cost for the infrastructure built from Oxchuc to Abasolo includes approximately 23 dollars (389 pesos) paid to Telmex, the Internet Service Provider; 21 dollars

(350 pesos) paid to the owner of the land in Lomen Loon; 6 dollars (100 pesos) paid for electricity in that place; 15 dollars (250 pesos) paid to the owner of the land in Cipres; and 3.5 dollars (60 pesos) for electricity in that place. The rent of the rooftop of a house in Oxchuc is paid through an agreement in which the household can have access to the internet for free. So the total monthly cost is 68.5 dollars (1.149 pesos).

Mariano Gómez explains that such costs do not include the installation and maintenance required to keep the network operating, which include the necessity of a car, oil, and time from him and others from the Collective as well as other indirect costs. Also, it does not include eventual charges asked by communities to allow for the installation of antennas. Lomen Loon authorities, for example, charged the collective approximately 470.50 dollars (8.000 pesos) for that.

121

The Colectivo Ik' Ta K'op charges different amounts for the service, depending on the involvement that people want to have with the project. They charge approximately 12 dollars

(200 pesos) per month to the ones that want to be just consumers, without further responsibilities. On the other hand, the ones who support maintenance of the towers and antennas, or collaborate in other Collective’s projects, such as the communitarian radio, can pay approximately 5–6 dollars (80–100 pesos), forming a kind of “distributed governance” (De

Filippi & Tréguer, 2016, p. 5) in the community. Finally, families can have the service for free if they affirm they cannot pay for it. The project’s financial management is very limited. There is no robust control of how many people have antennas, or how many pay each amount, or how much the Collective receives per month either, indicating that this is not a priority. Project management is as yet undeveloped, while the network and the number of members increase. All these characteristics distinguish the Collective from a for-profit business, and enable it to be recognized by international organizations as a Community Internet Service Provider (CISP).

5.1. Community Internet Service Provider’s Interconnection to the Larger Internet

Similar to Guelatao de Juárez, the point-to-point internet that started to be developed in

San Martín Abasolo in 2013 was a solution to replace the unstable and low-speed internet previously available. In Abasolo, the only option in the 2000s was a satellite to bring internet to that region. The signal now comes through a mobile network from a municipality where there is already an ISP offering the service. The Collective contracts the residential service from Telmex, in Oxchuc, at the household of an acquaintance. From an antenna that they installed in the house’s rooftop in Oxchuc, which is about 19 Km far away from Abasolo by road, the signal 122

travels through a wireless network to a mount in Oxchuc where there are two antennas—the receptive one points to Oxchuc and the emitter one points to Abasolo. There, a similar arrangement is set in another mount, from where the signal goes to the antenna installed at a house in Abasolo to be finally distributed to the other households in the community.

Behind this infrastructure, the Collective made decisions about where they would buy the internet signal. Such decisions are commonly based on convenience and efficiency—for instance, how far is the mount that can allow the signal to be well distributed in a certain locality, the availability of electricity in that place, the internet speed available to be contracted, etc., showing once more how technologies and the nature work as important actors shaping connectivity. The choice for a network design that situates the local network as a client of

Telmex services, on the other hand, reflects the control and power that local networks delegate to commercial ISPs in the moment of getting access to the internet. As some authors have already noted, “the need for uplink leads to the emergence of new bottlenecks that replicate the problems that community networks aimed to address in the first place” (De Filippi & Tréguer, 2016, p. 7).

Infrastructuring is always constrained by previous paths. In a large technical system, the design of interconnection infrastructure is critical to render explicit the exercise of autonomy that these local networks have.

Although these indigenous communities are implementing innovative connectivity strategies, the interconnection procedures clearly show their dependency on services offered by commercial ISPs. This not only exposes the undisputable prevalence of private actors in the governance of internet resources (DeNardis, 2010), but also raises questions about the role of internet infrastructure in the realization of communities’ values related to autonomy and sharing.

123

When it comes to content, more deeply problematized in this Case 3 in Abasolo as will be seen next, people express concerns about prioritizing platforms that do not represent local knowledge versus the aspiration of strengthening the communities themselves.

5.2. How to Leverage Local Content

One of the computers at the Mariano’s cyber café is a server where the Communitarian

Intranet Yajnoptik is hosted, a project that was first developed by a teacher to organize content for his students. The idea was that everyone within the community could access that local content for free, and in the future, the expectation is that the intranet comes also with a cloud space, for users to save their content, and a local social network, where users can communicate among themselves without using the larger internet. By the time of this writing, the intranet has a range of educational content, including a library of books from Latin America and other parts of the world, movies and documentaries, a version of Wikipedia, courses such as Khan Academy in

Spanish, among others, cached in the community’s server. Despite the fact that the community in general is not aware yet about these resources, everyone with a mobile device in the area can access them. So far, the content available has been centrally produced, and in the future Mariano expects that anyone from the community can upload information.

Abasolo’s connectivity project uses internet infrastructure to provide the intranet content to the whole community, going beyond the idea of “internet access.” Technically, the radio antenna that receives the internet signal in the community is connected to a switch, in which both a Wi-Fi router and the Wi-Fi management software server are connected. To this computer, another switch connects the intranet server and the antennas that will be distributed in the 124

community. This is an interconnection arrangement that keeps the intranet content local while allowing users to access it, independent of whether or not they are online.

Mariano envisions each indigenous community having a server in the future to build its own intranet and its own cloud service, enabling anyone to share material with others locally. It is the concept of access to knowledge applied in a context where the access to internet is restricted and where autonomy and self-sustainability are valuable. There is also the idea of having messenger services hosted on the local server, so users can communicate among themselves freely and locally, as there are no cellphone operators and SMS services available in the area. And although there are plenty of messenger services online, access to an intranet is not restricted by the broadband availability. As a CISP, the collective has more control over the server capacity and the amount of content that can be exchanged locally in its own network.

In Abasolo, one can see the conception of a local ecosystem that, on a minor scale, reflects the viability of services that provide opportunities not only for users to create and share content, but also for collaborations with other organizations that can help to leverage that production. This is when one can see the rise of a communitarian sense of internet governance in the context of indigenous communities. As Mariano Gómez notes:

If we want to do internet governance, we do not just have to have infrastructure, our antennas, our towers and links. We have to have the logical part, software design. And not only. We also have to produce our own content, our own videos. The Collective Ik'Ta K'op that is dedicated to the deployment of networks is not going to start making videos. We do not know how to make videos. But there are other organizations that do. In an organization, I cannot rule the internet by myself. I need several arms and supports to make real internet governance (own translation).

The experience provided by the Colectivo Ik' Ta K'op and the vision that they have for the future, even if yet to materialize, exemplifies the most complete ecosystem of shared network that we can see among the cases studied. It is aligned with the vision of other interviewees, such 125

as Peter Bloom, the founder of Rizhomatica, who are concerned about the trend of building infrastructure in the communities to strengthen the power of big corporations without empowering local people and their own knowledge:

It is interesting to see how the same [content] companies are getting into the internet access [business]. For them, the problem is that half of the world is not online. It's as if you had such a successful book company, that the only thing holding you back is that there are people who cannot read. Then you have to invest in literacy of the population so they can buy your books. These companies are at this point. The question is: who is going to build the access infrastructure and for whom? Why are we going to strive to build an infrastructure that 80% of the traffic will be from Facebook, Google and Microsoft? Another piece of this is missing (own translation).

Once local content and personal data are available on the network, who has the right to access it is a question to be raised as a way to make visible the “technopolitics,” or “the hybrid forms of power embedded in technological artifacts, systems, and practices.” (Hecht, 2011, p. 3), within internet infrastructuring experiences.

6. Security and Privacy

The concerns that one has nowadays regarding pervasive access from commercial ISPs should also be extended to other kinds of internet providers in the communities. Regarding the resources available, in the Guelatao municipality, there is a server which does the link distribution, assigns the Internet Protocol addresses and has a firewall that allows for the administrator to see who is online, who is using more broadband, and in some cases, which apps are being used. Facebook, WhatsApp or YouTube, for example, use standard ports, as Humberto

Morales explains, that allow for this identification, although encrypted content is protected. It is also possible to know the category and the websites accessed, which means that if a user is

126

accessing a porn network, administrators are able to know it. The use of this information is said to be restricted to create new rules on the server. For example, once a porn website is accessed through a public hotspot and the server informs it, the administrator blocks that website from then on, since in Guelatao, one has the right to see porn only if they pay for the internet. In public hotspots, it is not allowed.

In Abasolo, for the purpose of technology management, the Collective uses an open source Wi-Fi hotspot management software, Easy Hotspot, which generates the passwords automatically, and helps to manage each user’s time online. It also has the capability to show how many devices are connected, to disable a user, and to control broadband availability, among other functions. This kind of software—along with firewall applications for security, monitoring traffic and blocking unwanted content—is key to understand, at a basic level, the information access that a network operator has, and the tenuous line between technical and political functions that one has when intermediating connectivity. In Abasolo, where there are no firewall applications at the time of writing, Mariano Gómez defines the network as “free” and “neutral” because it allows “anyone” to access “whatever” content they want, limited only to broadband availability.

Analyzing the reality of “wireless community networks” (WCN) in Europe, defined as

“grassroots community networks, deployed at the local or regional level, managed by the community and for the community” (De Filippi & Tréguer, 2016, p. 4, emphases in the original), authors have found that “given the lack of a central authority regulating access to the network, it is in theory more difficult for anyone to assess the real identity of users connected to these networks” (De Filippi & Tréguer, 2016, p. 5). At the same time, there are no privacy terms of

127

service that the user agrees to, and no disclosure of what information is collected or available.

Entities controlling the networks do not divulge what information they view or how it is used.

While the interviewees of these initiatives are in consensus that personal data has never been used for purposes other than the one expected, as these experiences continue to expand, interest for data will certainly impose governance challenges for these shared networks. In

Mexico, organizations such as the Article 19, the Red en Defensa de los Derechos Digitales

(R3D) and Rancho Electrónico have actively advocated for the right for privacy and freedom of speech of Mexican people. How these discussions will be adapted and addressed by indigenous communities and their autonomous political systems is yet an open and embryonic discussion.

7. Ways of Approaching Shared Networks

I adopt the term shared networks instead of “community networks,” to emphasize the internet first (also known as last) mile signal sharing that characterizes all the cases examined here, independent of being led by for-profit or not-for-profit organizations. It is also a way of avoiding a term embedded in multiple expectations as community networks (see for instance

DC3 (2017)). Nonetheless, shared networks should not imply that all networks have the same sharing-oriented principles. Although all initiatives here examined are sharing or allowing the share of a first mile access point signal, among for-profit organizations, specifically, signal sharing seems to be acceptable because it is unavoidable, as equipment and individual skills for deploying it are increasingly accessible, and the charged prices are motivation to split costs among neighbors.

128

In the field, which involves scholars and organizations that support connectivity projects, authors have seen “community networks,” “wireless community networks” or “grassroot community networks” as initiatives that represent a kind of redemption of values from the beginning of the internet (De Filippi & Tréguer, 2016; O’Flaherty, 2018). Sometimes, they are considered alternative solutions to areas where investments from the private sector are considered not viable:

At the Internet Society we are interested in promoting community initiatives in these unprofitable places. We can think of it as a return to the academic origins of the Internet, where everyone makes the effort to “reach” the Internet instead of waiting for the Internet to reach us. It is to extend to the construction of infrastructure, the cooperative and collaborative model that we use for the administration of other resources (such as domain names or IP addresses). (O’Flaherty, 2018, p. 238, own translation)

Other times, understanding these initiatives as “the revival of more decentralized, citizen- centric network architectures” (De Filippi & Tréguer, 2016, p. 4), authors go another direction, interpreting them as “a counter-power to currently established power structures or incumbents”

(De Filippi & Tréguer, 2016, p. 5).

Within social movement perspectives, there is also the understanding that such “grassroot community networks” shape a “signal sharing movement” in Europe, whose participants are

“Not only (…) resisting the dominant corporate culture, they are creating a new culture”

(Shaffer, 2013, p. 247).

Finally, it is worth citing the coordinator of the Dynamic Coalition on Community

Connectivity (DC3), part of the Internet Governance Forum (IGF), who suggests that:

The emergence of community networks (…) [comes] with the objective of truly empowering the disconnected, allowing individuals and communities to self-determine, enjoying all the opportunities that connectivity can offer (…), at the same time that connected individuals contribute to the generation of even more services, contents and opportunities (Belli, 2018, p. 209).

129

From a local perspective, people who are building the networks, successfully connecting their communities and solving the problem of invisibility of their territories to the market and the state, express proudness about their communication achievements, reached through the mobilization of their own resources. While in Santa María Yaviche, these feelings are less prominent, as the community contracts the service from a regional ISP, in Guelatao, for

Humberto Morales, the internet quality improvement and the better costs offered for the population has represented a way to overcome the incumbent’s decisions of ignoring their requests. In Abasolo, the use of the spectrum to enable the internet means, above all, that the community is legitimately using resources that are part of its own territory to enable communication. As Mariano Gómez says:

There are things we already knew about the internet, but we did not know the name. The spectrum, we call air (own translation).

Remarkably, the history of indigenous populations in Mexico, and more visibly in

Chiapas and Oaxaca where the present fieldwork took place, is marked by tensions against cultural assimilation processes (Muñoz, 2005), leading indigenous authors to defend their diversity, autonomy, and the indigenous ways of living, known in Zapoteco communities as

“comunalidad” (Martínez Luna, 2010).

In the same direction, scholars who are also participants of other communicative experiences in these same regions, such as “community radios,” and “community cell phone networks,” explain the projects’ rationale or their results based on values related to autonomy, emancipation and self-sustainability (Baca-Feldman et al., 2018; Bloom, 2015; Huerta, 2018;

Parra, 2015).

130

While not ignoring these meanings associated with autonomy and means of community recognition that the actors give to the phenomenon, as well as the counter-hegemonic intentions and potentials of indigenous communities’ practices, I propose that the conflicts that emerge from the moment where the local gets access to the global internet infrastructure not be dismissed. A visible and regretted conflict is that these community networks primarily provide access to large content providers such as Facebook, Google’s YouTube, Netflix, and others.

Another conflict that has not been voiced—and I want to address—is the one that occurs at the level of network interconnection.

8. Values and Power in the Interconnection Arrangements

These shared networks at a first mile level are very particular when compared to the national ISPs companies’ networks with which they are connected, for access to the larger internet. First, they are not autonomous systems, with the technical capacity of exchanging traffic in interconnection points. Second, the Guelatao municipality and the Collective in Abasolo do not have the economic capacity to do wholesale internet transit purchase with distributors of carriers.

As already seen, one of the ways that these indigenous communities are building their interconnection to the larger internet is purchasing the service from larger ISPs whose services are available in other locations. The communities create their local interconnection points in a private household in areas covered by such companies, and from there they install antennas all the way down to their own territories, where other receptive devices are set up in households and public spaces. 131

In fact, from the standpoint of the interconnection strategies, it is possible to say that what these indigenous communities are infrastructuring is a new network that both includes them by connecting the people that are behind the local infrastructure to the global internet, and expands the reach of big and commercial ISPs’ networks to their localities.

In her studies on interlink infrastructure applied to electricity systems, Jane Summerton calls attention to the role of entrepreneurs in an actor network, who will “seek to expand their power and increase their control over other actors, entities and resources that are crucial to reaching their own goals” (Summerton, 1999, p. 95). Nevertheless, in ANT, identities, goals and the limits of entities are defined during action and not independently, in a process named

“translation” (Callon, 1984).

In Cases 2 and 3 examined above, the Guelatao municipality and the Colectivo Ik' Ta

K'op take action, mobilizing many resources and responding to the limitations imposed by technologies’ affordances to enable connectivity, despite ISPs positions in ignoring their requests. Moreover, they establish shared networks by design that disregard the restrictions imposed by ISPs’ policies on signal sharing. For instance, Telmex’s terms and conditions explicitly state that “THE CONSUMER acknowledges and accepts that the SERVICE is of a residential character for use in the household, so that THE CONSUMER cannot commercialize, sell or resell the internet service.” (Telmex, 2016, p. 4, emphases in the original, own translation). izzi forbids not only internet signal sharing for commercial purposes, but for any other (izzi, 2014).

These facts show the efforts of Tseltal and Zapoteco communities who, with the goal of establishing communication with the internet, signify the commercial ISPs services and the

132

internet infrastructuring as a means to get there—which also happens in Santa María Yaviche hiring the services from a regional ISP. However, the question of power and control remains open. Considering that this is a translation process “by which the social and natural worlds progressively take form,” by assumption, during actors’ interaction, “The result is a situation in which certain entities control others” (Callon, 1984, p. 224).

Looking at the materiality of the infrastructure built, the shared networks do not have substantial control of their internet communication. Instead, they are susceptible to the ISPs economic power and legal prerogatives materialized in their terms of services, as the companies can shut down the internet at any time. Additionally, in the position of ISPs’ consumers, Tseltal and Zapoteco communities are directly subject to the companies’ policies and possible deep packet inspection, a procedure that ISPs in general have taken, arguing the need for network management that leads to decisions over users’ traffic for commercial purposes. Monitoring, slowing down, blocking the sharing of content through peer-to-peer file-sharing platforms are some examples (Bendrath & Mueller, 2011). Furthermore, while by design, shared networks do not allow a direct association between the consumer who contracted the service and internet behavior—as access will be originated in different machines and by different people—device profiling is still possible as they have unique local area network identifiers, such as media access control (MAC) addresses. And as all the community access will traverse a unique residential connection, this tends to be a collective profiling.

Despite the communities’ laudable accomplishments, improving their territories’ connectivity, or providing access for the first time, the infrastructure design constrains self- sustainability values of these shared networks. As Jane Summerton synthesizes, “In actor-

133

networks, control by the dominant actor is accompanied by the loss of autonomy by all others”

(Summerton, 1999, p. 96).

In view of how important autonomy, emancipation, recognition, and self-sustainability are within these communities and their communication projects, I propose understanding local internet infrastructuring as a moment that calls for recognition of the internet infrastructure as part of the network of relations that is deployed around values.

From the standpoint of the internet interconnection arrangements, let’s assume that values are “hypotheses,” (JafariNaimi et al., 2015) and emerge in context, in face of situations that require actions, as framed by John Dewey, on whose work these authors build. In the cases above, the lack of connectivity leads people to claim the right to communication in action, and this is one of the values that guides infrastructuring towards internet access. The interconnection arrangements with the larger internet can be understood then as “values seams” (Shilton, 2018), where tensions between communication values and autonomy-related values take place. As

Batya Friedman and colleagues put it, “[A] given technology is more suitable for certain activities and more readily supports certain values while rendering other activities and values more difficult to realize” (Friedman et al., 2006, p. 4).

Therefore, it is useful to not think separately of humans and their values on the one side, and infrastructure on the other, as if the first ones were independently driven, as some of the current analyses of “community networks” surmise. The tensions here exposed are better understood in terms of “programs of action,” where artifacts are actively constitutive of societies’ morality (Latour, 2008). Internet infrastructure influences the course of the facts, and

134

while enacting communication values, its interconnection arrangements may constrain autonomy, emancipation and self-sustainability.

9. Final Considerations

Globalization studies have largely discussed the interactions between local and global spheres. They highlight that “Studying the global, then, entails not only a focus on that which is explicitly global in scale, but also a focus on locally scaled practices and conditions articulated with global dynamics (…)” (Sassen, 2003, p. 3). Looking at the local helps to shift our attention to a generally dismissed topic in global internet governance, that internet design is far from being exclusively in the hands of specialized and technical people. Instead, it is a construction taking place from and within many different entities. For Arjun Appadurai, who discusses the relation between global, local and the “social life of design,” the local is constantly being designed. As so, he shrewdly concludes that “design is only partly a specialist activity, (…) and is better seen as a fundamental human capacity and a primary source of social order” (Appadurai, 2013, p.

254).

In this context, the pragmatic need for indigenous populations to communicate fulfills the rationale for regional Internet Service Providers, municipalities and social organizations to mobilize technological resources in search of connectivity. They engage in internet infrastructuring towards a digital citizenship neglected by the state, in a context where the internet connectivity rights stated in the Constitution are unfulfilled.

Internet infrastructure, in this scenario, gains numerous codesigners at a local level, surpassing its invisibility to common people and becoming subject to infrastructuring where it is 135

absent in the communities. Such a task is not exclusively dependent on human labor but involves natural and technological resources including the air/spectrum, hills, home rooftops, antennas, routers, towers, tree towers, and internet signal made available by ISPs, who all have a role in the outcomes, and introduce path dependencies through their affordances and characteristics.

If we understand that design adds context to things and “regulates” the numerous possibilities that things can have (Appadurai, 2013, p. 263), not only are we led to think that, as internet codesigners, regional ISPs, Tseltal and Zapoteco communities are adding context to internet infrastructure; they are also constrained by its previous deployment, as “Infrastructure does not grow de novo” (Star & Ruhleder, 2015, p. 381). This is when local efforts to adapt a large technical system face limitation and values go into tension.

Interconnection arrangements that allow the local to become global are key places where these tensions can be explicit. The arrangements people have chosen in these shared networks elucidate the gulf between economic and infrastructure characteristics of their networks and of the traditional ISPs present in other towns. When local shared networks interconnect and shape an actor network with larger ISPs, power imbalance becomes clear and local autonomy that has guided the infrastructuring process is impaired in favor of another key value for them: communication. Other infrastructure solutions are necessary to reconcile these values.

At a global scale, public interconnection, or internet exchange points, has been used by commercial networks and non-commercial ones to facilitate internet connectivity. The next chapter will focus on the formation of the first IXP in Mexico City and the power dynamics that emerge from it that makes the IXP a distant infrastructure solution for the shared networks analyzed.

136

CHAPTER 3

THE POLITICS OF AN INTERNET NODE: BUILDING THE FIRST INTERNET EXCHANGE POINT IN MEXICO32

When people say of technologies that they are neither good nor bad, they forget to add: nor neutral.

—Bruno Latour, An Inquiry into Modes of Existence (p. 219)

This chapter examines the formation of the first internet exchange point (IXP) in Mexico and proposes to define an IXP as a network of relationships; it shows how organizations, individuals, documents, laws, and technology artifacts are significant actors in the making of the IXP; it explores the ways that social, political and economic goals are delegated to the IXP’s technical functions, and the limitations of the imaginary related to IXPs’ social benefits; it argues that the lack of convergence among these actors prevents the project from succeeding, while regulatory documents work as the supporters of a fragile equilibrium to keep the project ongoing.

1. Introduction

In May of 2014, a cadre of companies, policymakers, and journalists convened at an event for launching a new part of the internet architecture in Mexico—an internet exchange point

(IXP). As seen in Chapter 1, an IXP can be initially understood as an internet node, a physical

32 A version of this chapter was presented at the TPRC46 (46th Research Conference on Communication, Information and Internet Policy) under the title “Internet Node as a Network of Relationships: Sociotechnical Aspects of an Internet Exchange Point.” I would like to thank the anonymous reviewers for their comments and recognition of the paper as one of the best student papers of the conference. 137

facility where different networks can interconnect and make private agreements for the purpose of optimizing their respective resources to exchange traffic and routes on the internet. Networks, in this context, are mainly Internet Service Providers (ISPs) (e.g. Comcast, AT&T, Telmex) and content providers (e.g. Google, Facebook, Amazon), but can be also banks, universities and other organizations which have an autonomous system number (ASN) to uniquely identify their networks on the “network of networks,” or the internet. Commercial agreements between networks shape the way that they will interconnect with each other, if by “peering” or by

“transit,” as discussed in Chapter 1, and IXPs facilitate these agreements as it becomes a convenient point for networks to meet.

When a country does not have an IXP, local networks may establish private agreements in other locations if they agree that there are mutual benefits. They also tend to exchange traffic abroad, as a way to transmit as much data as possible through less expensive options than transit agreements, as explained in Chapter 1. Specifically in Latin America and the Caribbean, not more than a third of the nations had IXPs in 2016 (Galperín, 2016), and it is estimated that a great amount of data goes from the Latin American countries to the United States (Agudelo et al., 2014), even when source and destination are from that same region. This is known as boomerang traffic, detour or trombone, and makes data transmission more expensive and the internet subject to greater latency.

In Mexico, at the time of writing, the IXP under study is responsible for a low amount of internet traffic, with a speed of 10 to 20 Gigabits per second (Gbps) according to interviewees.

For a rough comparison, as information about the internet traffic per country is privatized and not publicly available, the main IXP in Latin America, located in Sao Paulo, Brazil, has an

138

average of more than 2 Terabits per second (Tbps). As populations and internet penetration in both countries are different, to contextualize these numbers it helps to know that traffic originated in Mexico is responsible for 1.5% of the global web traffic while Brazil, located in the same region, responds to 3.5% (Akamai, 2018). Another metric to contextualize the low volume of traffic passing through the first Mexican IXP is the number of autonomous systems in the country connected to that facility: of the 366 autonomous system numbers assigned within

Mexico, fewer than 10 are connected to the IXP. Together, these sources can be considered an indication that the internet traffic in the country continues to pass through bilateral connections in private facilities that precede the relatively new available IXP in Mexico City.

Nonetheless, with regard to the launching of an IXP in Mexico, Carlos Casasús, who is the president of the committee formed to coordinate the new facility, mentioned to a journalist some benefits that would justify the implementation of the first IXP in the country (Rivera,

2014). His considerations encompass four key issues:

1. Leveraging the quality of the internet, through the “decrease of latency between

connections” and the “improvement of the internet traffic”;

2. Strengthening sovereignty, through avoiding unnecessary international routes,

“enriching the country’s technological infrastructure,” enabling the country to join

others “that are at the forefront of technology”;

3. Leveraging market competition, helping to establish “a healthier competition

among telecommunications operators,” and “attract more foreign investment”;

and

139

4. Generating social benefits, “narrowing the digital divide by making the internet

more accessible to more people,” and “encouraging further development of

national content online.”

While these reasons reflect local motivations, they incorporate components of a prevailing dialogue among international organizations. Many different agencies, including the

Organisation for Economic Co-operation and Development (OECD), the Inter-American

Development Bank (IDB) and the World Bank, have produced reports on broadband development, emphasizing the role of IXPs in improving connectivity rates in “developing” countries (Agudelo et al., 2014; Blackman & Srivastava, 2011; Intven et al., 2000; OECD &

IDB, 2016; Weller & Woodcock, 2013).

By unveiling a hidden technology underlying the internet architecture, what this chapter elucidates is that the expectations about the first IXP in Mexico are based on assumptions that depend on different sociotechnical processes and actors intertwined, in line with the concept of

“imagined affordances” (Nagy & Neff, 2015). In action, these imagined attributes merge their characteristics with who handles them, supporting new actions that emerge in a process named

“translation” (Latour, 2002)—only conceivable if object and subject are considered altogether.

In this chapter, I analyze the incomplete realization of such a translation process in the case of the Mexican IXP, or the reasons for both the expectations of some groups involved in broadband discussions and in the deployment of the IXP to be frustrated and a lost link between the artifact and the social benefits to be created.

Case studies focusing on single IXPs have been used to examine the London IXP – LINX

(D’Ignazio & Giovannetti, 2009), the Slovak Internet eXchange – SIX (Restrepo & Stanojevic,

140

2012), the Bolivia IXP (PIT-Bolivia) (Carisimo et al., 2015; Galperín, 2016) and the Brazilian

IXP ecosystem (IX.br) (Brito et al., 2016) with an engineering lens or social sciences lens informed by quantitative methods and complementary in-depth interviews. The present chapter is original in its approach for examining an IXP from the Science and Technology Studies (STS) standpoint for the first time, to the best of this author’s knowledge.

I apply this approach to the study of IXPs using three principles presented by Michel

Callon (1984): agnosticism – to be impartial with the parts of a controversy; symmetry – to analyze different perspectives with the same lens; and free association – to break the divide between society and technological artifacts.

2. Making an IXP

In 2012, the OECD released an influential report on Mexico, one of its few member- countries from the Global South, stating that “The welfare loss attributed to the dysfunctional

Mexican telecommunication sector is estimated at USD 129.2 billion (2005-2009) or 1.8% GDP per annum” (OECD, 2012, p. 9). Among its recommendations, the report stated that the telecommunications regulator, the Instituto Federal de Telecomunicaciones (IFT) (Federal

Telecommunications Institute), should have the power to impose regulations and sanctions to leverage competition. With regard to infrastructure specifically, the report says that “The inability to mandate, or at least set out, reasonable conditions for infrastructure sharing is arguably one of the main bottlenecks that prevent competition” (OECD, 2012, p. 12). Since then, the report has been a respected voice in policymakers’ circles discussing infrastructure-sharing projects and the intensification of asymmetric regulation applied to the preponderant economic 141

agent, Telmex. This company used to be a public company and its license to the private sector, primarily focused on telecommunication services, had already stablished “the obligation to both have open architecture and interconnect its networks” (Álvarez, 2018, p. 42, own translation).

Carlos Casasús’ story of conversations about creating an internet exchange point in the country at the regulatory agency is an example of these discussions applied to the internet specifically:

We were already talking about having an IXP. I was the chairman of COFETEL's [Comisión Federal de Telecomunicaciones] Advisory Board [currently IFT]. I had a meeting with the COFETEL’s president [Mony Sacha de Swaan] and I said ‘Why do not we do that? It is an OECD recommendation.’ He said: ‘Do you think we can do that? We have been working for many years…’ So, we managed to get [some] partners to start.33

Casasús is known for his efforts within the not-for-profit organization Corporación

Universitaria para el Desarrollo de Internet (CUDI), whose goal is to congregate and escalate resources among higher education institutions in Mexico.34 It is in this context that he and colleagues thought about building an IXP first in the beginning of the 2000s to improve universities’ internet connectivity, keep the country’s content local and decrease dependence on the United States’ infrastructure. Hans Ludwing Reyes Chávez, one of the engineers who work for CUDI and who is currently responsible for the IXP in México, remembers that at that time:

“[The idea] did not prosper because there were not enough fiber networks to do it.” In fact, network interconnection depends on numerous infrastructure resources including optical fiber and broadband links.

33 This and other verbatim quotes come from interviews with the author in Spanish and my own translations. 34 Previously to this role, he was the Financial Director of Telmex, when it was a state company, and worked in the front of the Federal Law of Telecommunications discussions, approved in 1995 (www.diputados.gob.mx/LeyesBiblio/abro/lftel/LFTel_abro.doc). He was then the first COFETEL president in 1996, the regulatory agency that since 2014 is called IFT. 142

According to Carlos Casasús, an inspiration for CUDI and the IXP project has been the

Rede Nacional de Ensino e Pesquisa (RNP) (Brazilian National Research and Educational

Network), a network of universities in Brazil whose goal is to integrate academic institutions with the support of a backbone fiber network running since 1992. At the time of writing, RNP has access points in all 27 Brazilian states, facilitating the interconnection of networks in different regions, and serving as points of interconnection of some IXPs within the country.

Unlike RNP, though, CUDI does not have a fiber network in Mexico. The organization depends on an agreement between the Secretaría de Comunicaciones y Transportes (SCT) (Ministry of

Communications and Transportation) and the Comisión Federal de Electricidad (CFE) (Federal

Electricity Commission), which interconnects approximately 40 universities, but constantly presents technical problems, according to the interviewees.

This is an important context to understand, that the first IXP initiative in Mexico was led by an educational organization with clear purposes, but devoid of internet infrastructure resources. In 2014, CUDI, and more specifically its president, put together five companies to start the exchange point in Mexico City: Kio Networks, Megacable, Nextel, redIT, and

Transtelco. These organizations constituted the IXP’s founding partners, which envisioned some benefits for themselves, including sharing infrastructure and exchanging traffic among the parties, and in the case of Kio Networks—a prominent data center within the country—the opportunity to become the host of new networks. Interestingly, the group of the IXP founders does not comprise Telmex, the telco incumbent, and other academic institutions than CUDI, which would be required to have autonomous system numbers to interconnect, and is reported to have difficulties in receiving ASNs from NIC Mexico.

143

While the participation of a player like Telmex cannot guarantee the success of an IXP,

Telmex competitors and the IXP founders defend that it is a crucial contributor to it, given that

Telmex not only has the biggest number of clients, concentrating 57.7% of the internet market,35 but it also has the largest infrastructure to reach different parts of the country, with more than

190,000 km of optic fiber according to the company’s blog (Telmex, n.d.). For instance, an ISP, which needs to deliver data packets in places where its own optic fiber mesh does not reach, has two possibilities: buy transit or do peering with another company to deliver it. However, an incumbent agent has very few incentives to share its own infrastructure and peer with potential competitors. Economically, it can be more advantageous to sell transit to some ISPs than to peer with them. In Mexico, due to its reach, Telmex would be one of the most likely companies from which this supposed ISP would buy transit. Thus, for Telmex, it is reasonable to think that an internet exchange point would likely reduce its clients and would not benefit its business.

Taking part on this controversy, and guided by the purpose of leveraging competition in the country, two months after the beginning of the IXP operation, the law that marks the reform in the telecommunications sector was passed determining asymmetrical regulation to the preponderant agent, Telmex. It says that the company should: “Have a physical presence in the

Internet exchange points in the national territory, as well as to enter into agreements that allow

Internet service providers the internal exchange of traffic in a more efficient and less expensive way according to the terms that the Institute define” (Mexico, 2014, own translation). As of the time of writing, though, Telmex was not yet an IXP member, but the expectations were that it

35 The other big players are Grupo Televisa, with 21.5% of the market, and Megacable-MCM, with 13.5% (IFT, 2017). 144

would happen soon, if the company does not appeal to the guidelines issued to enforce the law in

2017.

According to these guidelines, the preponderant agent or the agent with substantial market power “must establish Connectivity through the deployment of fiber optic links to IXPs that request it, and where there is at least one Internet Service Provider with which [it] does not have a traffic exchange agreement [peering agreement] (…).” (Mexico, 2017, Cap. III, own translation). Moreover, it “(…) must advertise the Routes of [its] clients and accept the Routes of the ISP members of the IXP. The Routes must be kept constantly updated in the Routing Table”

(Mexico, 2017, Cap. III, own translation). With such rules, all the costs for the incumbent to be connected to any IXP are supposed to be covered by the company. Furthermore, by establishing peering agreements with any ISP connected to an IXP within the country, Telmex will lose the possibility of selling transit to its competitors when they want to reach Telmex’s own networks and Telmex clients’ networks.

This type of regulation requiring interconnection, while it has reflected significant lobbying from CUDI, has not been received unanimously among players in the market and specialists. In the illustrative opinion of a content provider representative, who is responsible for interconnection issues at a company that already maintains private peering agreements with

Telmex, he notes that an IXP is useless in a market where there is a low level of competition, and not an ISP ecosystem to benefit from interconnecting publicly at an exchange point. He defines the Mexican IXP as a “party where all the guests already have relationships with each other,” so pay for a “ticket” to participate in such party is a waste of money. In other words, for this interconnection specialist, in a market like Mexico, the equipment necessary to build an IXP and

145

the structure necessary to maintain it becomes costly, and will not solve the competition problem by itself:

A switch helps the small players, but if they do not exist (…) an IXP will not generate small players.

In this vein, he sees the regulation to require a player to participate at an IXP as an unwelcome interference: “When there are no commercial reasons [to interconnect], one makes the law,” he says.

On the other side, IXP defenders and pro-regulation actors shift the focus to the challenges faced by both small players that do not have the market power for interconnecting directly via private facilities, or other Telmex medium-size competitors, which depend exclusively on transit services because of Telmex business strategies and policies. It is not uncommon that a Telmex competitor who wants to reach Telmex networks needs to send its traffic to an IXP in the United States, where global internet networks—also known as Tier-1 networks—which keep agreements with Telmex will redirect such traffic to return to Mexico.

IXP defenders and pro-regulation actors will say that this boomerang route raises cost issues for the companies, internet quality issues for the users and sovereignty issues for the country, the later primarily related with foreign surveillance concerns.

Julio César Bravo, an incumbent competitor representative whose company is one of the

IXP founders, believes that there are viable business opportunities to raise in an IXP in México, but Telmex needs to be part of it to make it attractive to Content Delivery Networks (CDNs).

CDNs are services provided by third-party companies that cache highly accessed web content to make it easier and quicker for users to reach. Big content providers (e.g. Google, Netflix) have also developed their own cache infrastructure and are known by IXP operators as CDNs as well.

146

All these content players have an interest in becoming a member of an IXP if a great amount of traffic is expected to circulate through its facilities. Although Julio César Bravo would agree that the IXP is currently a party with guests that are already linked among them, his company accepted to be an IXP founder based on future business perspectives, such as providing connection for the IXP to the United States. He admits:

In the end it is business (…) There is no altruistic issue. Everything is totally and completely business.

Thus, for the IXP team and participants, the state regulation to require the incumbent to be part of the exchange point is positive, and generates expectations that other important players will interconnect to the exchange point in the near future. This is in line with scholarship on IXPs in the region, although “heavy-handed government regulation of IXP operations is ill-advised,”

(Galperín, 2016, p. 639) due to the flexibility necessary in the internet interconnection realm. On the other side, consequential players, including the incumbent and the ones that already have interconnection agreements with it, such as big content providers, do not see benefits from connecting to an IXP in the present conditions. In fact, although the IXP has already been working for some years, its outcomes have not been measured or made public, which generates critics:

I have no elements to know if I can trust the IXP operator or not. In theory, yes, because I'm in a university and I have to rely on CUDI, right? But I do not even know where IXPs’ performance measures are, if I do not have numbers I cannot have confidence

This is what Luis Miguel Martínez Cervantes says, a university professor and also the

Internet Society Chair in Mexico, an organization that has supported the creation of IXPs around the world.

147

Luis Martínez argues further that building an IXP in Mexico at that moment was “a political and not a technical decision,” meaning that the IXP was a government response to the

OECD report agenda, while his academic colleague, Judith Mariscal, a university professor and specialist in telecom and digital divide issues (Flores-Roux et al., 2009; Galperin & Mariscal,

2016), argues that the IXP was Carlos Casasús’ and CUDI’s agenda, indicating lack of involvement in the discussion. Clearly, CUDI’s IXP lobby was directed to government and some companies and did not incorporate other academics and civil society organizations in its process.

To finalize this examination session, it is important to explain the governance and design of the IXP, here understood as two sides of the same phenomenon (DeNardis, 2014; Musiani,

2013). The IXP governance is under the auspices of the not-for-profit organization Consorcio para el Intercambio de Tráfico de Internet (CITI A.C.) (Consortium of Internet Exchange

Traffic), which is led by the CUDI president, Carlos Casasús, and complies with the partner organizations of the IXP that meet every three months. As of the time of writing, organizations connected to the IXP are Akamai, Cloudfare, CUDI, Enlace TPE (TotalPlay Empresarial),

Google, KIO Networks, NIC Mexico, Megacable, y Transtelco. Interestingly, some companies that were connected to the IXP in its beginning are no longer connected. This happens because there is a merging trend among businesses (e.g. AT&T bought Nextel Mexico and KIO

Networks bought redIT), an expression of technology convergence that may reduce the number of IXP participants in a small market.

KIO Networks is the company that owns the data center which hosts the IXP’s equipment, being responsible for the co-location and the building infrastructure—electricity,

148

cooling and security. It has an important role in IXP governance, once its policies are crucial in the design of the IXP and its geographic location.

To be part of the IXP consortium, the organizations need to pay $810 or $2,430 monthly to have a port of 1Gbps or 10 Gbps, respectively, but companies such as Content Delivery

Networks may negotiate these terms due to the perceived importance of caching highly accessed content locally for the economy of IXP participants. To be connected to the IXP, a network— owned by a company, a community or the government—needs to be an autonomous system (or systems), which means having an autonomous system number assigned by NIC Mexico, and to be physically connected to the IXP in Santa Fé, Mexico City, where the KIO Networks data center is located. If an interested network is already based in this data center, it will purchase a

“cross-connection” service from KIO to have its cables connected to the IXP. If this is not the case, a point-to-point link from the company headquarters to the IXP is necessary. In this scenario, one of the challenges is that the usage cost for local fiber lines is expensive and wireless lines are not abundant in the country, contributing to preventing IXP attractiveness. As

Luis Martínez exemplifies:

What happens is that for [my network] to arrive from a town 10 km from the IXP, I have to use the Telmex network. And in this case, I find it cheaper to use the Telmex internet service than what the IXP is going to give me. Because what Telmex will charge [for a fiber line] to take me to IXP is going to be more than what Telmex will charge to provide me the internet service without having to go to the IXP.

Part of this scenario is due to the access that the incumbent has to passive infrastructure throughout the country, including antennas, posts, and right-of-way—the legal possibility of passing cables through public spaces. Because Telmex used to be a public company, it has kept better negotiations with supporting infrastructure historically. In contrast, small players have

149

more difficulty to have access to right-of-way according to interviews, making competition even more problematic.

3. The IXP as a Result of Interconnection Politics

The analysis that follows is based on the actors that stood out in the dynamics of the IXP formation: the OECD report, CUDI’s president, the telecommunication regulator, the telco incumbent, the telecommunications law and IFT guidelines, NIC Mexico, the fiber networks, the passive infrastructure (posts, optical fiber, right-of-way), big content providers/CDNs, incumbent competitors, global networks (Tier-1 network), civil society more broadly (including academics) and the core actor, the IXP, that from the narrative goes beyond its equipment—cabinet, switch, router, cables—, and includes the data center, the networks connected to it, and the governance consortium team. Independently of being human or non-human, actors are considered symmetrically, including individuals, networks, supporting infrastructure and documents, who have had an active role in the dynamics. In ANT, action is conceived not as an exclusive

“property of humans,” but as a result of a combination of agents or “actants,” including technical artifacts (Latour, 1999). Regarding documents, the very argument to consider them more than merely sources of information is that text transcends authors and their intentions. They can instigate actions and can “be considered as actors in their own right” (Prior, 2008, p. 822).

From a policy perspective, it stands out that infrastructure aimed at improving connectivity within a country, as it is the case of the IXP, is a result of numerous social, political and economic goals that are delegated to this artifact, in a translation process in which actors’ identities and characteristics are negotiated in relation to the others. Michel Callon (1984) 150

suggests four moments of observation to understand this translation development: problematization, interessement, enrolment and mobilization. These moments are not independent of each other, though, they are dynamics that can overlap.

3.1. Problematization

The problematization moment is when certain actors “establish themselves an obligatory passage point in the network of relationships they [are] building (…).” When they become

“indispensable in the network.” (Callon, 1984, p. 204). In the present case, this actor is the IXP, voiced by CUDI’s president who can be considered IXP’s “spokesman,” in Michel Callon’s terms.

The goal of building the first IXP of the country required CUDI’s president to negotiate with several actors. In this context, the OECD report worked as a catalyst for the interconnection facility formation once it recommended reducing market concentration. The IXP promises, echoed by CUDI’s president, conveyed this possibility, which was in accordance with the telecommunications regulator interests. The document worked as both a symbolic and material supporter for CUDI’s president to resort to it in his dialogues to enable a group of supporters.

Notably, even when the IXP was just a project, it was already an actor in terms of the outcomes expected in the market and in society. The question was if there would be enough support to physically build it. CUDI’s president starred the problematization phase, defending that it was the best moment for joining efforts to build an IXP, and that such a technological artifact was the best answer to address not only economic disparities in the market, but also social and political issues, including digital divide and national sovereignty. 151

3.2. Interessement

The interessement moment arises when the IXP project needs to attract enablers and distance them from other alternative responses to the existent problems. CUDI’s president defended that, once formed, an IXP would improve internet traffic and quality; avoid international routes and strength sovereignty; leverage market competition; narrow digital divide and encourage development of national content online. Interestingly, CUDI has for a long time been interested in improving Mexican universities’ connectivity, so sharing infrastructure in the

IXP was seen as an alternative to their difficulty in negotiating effective fiber networks connections given that the organization is devoid of an academic backbone network.

The regulator, IFT, heard CUDI’s president voice parallel to the OECD report repercussions and moved to delegate its policy goals to law and guidelines requiring the telco incumbent to participate in the incoming IXP. The law issued in 2014 worked as a guarantee for companies to invest and engage in the project, even if in a small number. The promise of making peering agreements and sharing infrastructure with the incumbent in the near future supported such private investments. The players interested were in unison, understanding that without the law, the telco incumbent would not integrate the project.

Advertised outcomes of the IXP showed technical, political and economic purposes completely intertwined. Beyond the government collaboration, they attracted companies interested in optimizing their costs and leveraging their profits based on the belief that in a certain period of time the IXP would deliver what had been promised, especially traffic exchange with Telmex. The specificity of the networks attracted to the IXP project is that they were at a disadvantage in the market in comparison with the incumbent infrastructure and the dependence 152

on Tier-1 networks in the United States to connect to the Telmex network. The regulator’s law and guidelines requiring Telmex to be part of the IXP give the reasons necessary for them to join the project, and more importantly, keep the project ongoing even after some years of no expected results. The law and the guidelines, which the telecommunications regulator issued to enforce the law, are key actors for keeping the IXP live in a fragile equilibrium.

Remarkably, companies read the expected outcomes with an economic lens. For instance, reducing the international traffic means saving money in traffic costs and decreasing latency, while CUDI and government would defend that it means strengthening Mexico sovereignty.

Thus, the association of broad social and political benefits to the IXP does not have the same significance or attractiveness for different actors involved.

3.3. Enrolment

Callon points out that “To describe enrolment is (…) to describe the group of multilateral negotiations, trials of strength and tricks that accompany the interessements and enable them to succeed” (Callon, 1984, p. 211). The IXP formation depended on actors not always visible and ready to support the project: a data center designed to securely host its equipment—servers, switches, routers, cables, fiber internet links—and autonomous system numbers. For the networks to interconnect using an IXP they need to negotiate with these actors; otherwise, they become barriers for networks to effectively be part of the IXP. For instance, the difficulty faced by some universities to be assigned an autonomous system number by the NIC Mexico has kept them apart from the IXP. Legal and economic constraints that restrict the offer of affordable fiber links to where the data center is based can reduce interest of regional networks based far from 153

Mexico City in connecting to that internet node, as well as induce them to continue buying internet from the incumbent as exemplified earlier by an interviewee. Additionally, to keep the

IXP equipment functional and sustain its colocation at the data center, IXP members are asked to pay a monthly contribution in dollars, which also becomes a barrier for small internet service providers. In the end, the design and governance of the data center are altogether crucial for IXP performance, not only for what they allow, but also for what they constrain. While the IXP itself is considered to be physically formed by a cabinet with switches, routers, servers and cables, it is in fact intertwined with the attributes of the data center where it is colocated, the networks that are successfully connected to it, and the ones that are not connected due to failed negotiations with other infrastructure actors marked by legal and economic barriers.

There are certain actors that were not involved in the formation of the Mexican IXP, although the social outcomes that the IXP spokesman advertised to attract supporters are of great interest to them. These include civil society groups who advocate for affordable internet and are responsible for building wireless shared networks in communities where internet service providers are not willing to serve, as well as academics who are important voices in the area of telecommunications and the digital divide. Considering that, despite CUDI’s president and the telecommunication regulator, the other active actors engaged in the formation of the IXP who voiced their interests do not mention concerns with the digital divide or with national sovereignty. Such promised outcomes seem to be primarily a rhetoric tool for the IXP spokesman, and not a mobilizer used to aggregate actors previously interested in these issues around the IXP.

154

Interestingly, in this case, IXPs become similar to other infrastructures in which beyond their technical functions, their form, “or the poetics of infrastructure” (Larkin, 2013, p. 329) shows political facets through the “imaginary” and the “fantasy” created around them.

Furthermore, as Cynthia Cockburn argues, if based on the way and by whom they were built, technologies are masculine and cannot be seen in a sexless mode (Cockburn, 1983), as a technology, the Mexican IXP cannot be seen as a non-commercial entity. As it is used to facilitate commercial agreements, even if led by a not-for-profit organization, its governance and design are commercially oriented by its members. In this scenario, modeling the IXP in the direction of social and political goals is thus unlikely to happen, once such outcomes are restricted to the desire of some actors.

3.4. Mobilization

Convergence and a certain level of consensus around a proposition mark the success of the mobilization moment. In the case of the IXP formation project and the actors that emerged in the dynamics, the mobilization results can be considered only partial. The project was formulated based not only on the affordances of an IXP—or what it can do—but on the successful translation of organizations’ goals into technology functions. Yet the fact is that after some years since the IXP formation, that didn’t happen. IXP development has maintained the interest of new networks in connecting to the first Mexican IXP limited, keeping the number of its members low. Lack of abundant and affordable links to connect networks in other regions to the data center shows the role of fiber networks and passive infrastructure as actors that constrain such interest. Companies that founded the IXP were acquired by other businesses and the IXP 155

stage was not enough to initially attract new big players to the project, such as AT&T who bought Nextel, a previous IXP member.

Public information about IXP performance is not available, but the reported IXP traffic in interviews is modest. Thus, there is no evidence that key promised outcomes, such as reducing international traffic and latency, leveraging competition and access to the internet for more people, have been addressed. Some academics are skeptical and still not engaged in the project.

On the other hand, it is not a trivial outcome that, despite all the frustrated expectations, the IXP in Mexico is still running while there are numerous defunct IXPs in the world.36 The mobilization moment that started with the formation of the group that would support the IXP formation, including the telecommunication regulator and some companies, has been continuously sustained. For this to happen, the most important actors in this scenario seem to still be the law and the guidelines that require the telco incumbent to adhere to the IXP. They generate the expectancy that after the Telmex connection, IXP traffic will exponentially increase, networks will not need to use Tier-1 networks in the United States to connect to the incumbent, and new networks will be attracted to the IXP, contributing to the likelihood that the IXP will prosper. Such results, however, are not a given. They are part of the infrastructure imaginary around the IXP and will depend on negotiations among actors when Telmex changes its position in the scenario. The regulatory documents thus support a dynamic equilibrium based on this imaginary that allows the project to continue.

36 A filter at the Packet Clearing House database (www.pch.net/ixp/dir) shows 118 defunct IXPs in the world as of July, 2019. 156

4. Final Considerations

In Chapter 1, some IXP definitions in internet network scholarship described the artifact in terms of its interconnection affordances, its position within internet layers, and the formal requirements for networks to interconnect (Chatzis et al., 2013; Euro-IX, 2015; Fanou, Valera, et al., 2017).

From a sociotechnical perspective, the evidences obtained with the present research enable understanding IXPs as a network of relationships that involves players with goals and functions that mesh to become an interconnection facility in the internet. Such networks of relationships are dynamic and are defined relative to each player in the scenario, which includes organizations—characterized by their design and governance—, individuals, documents, laws, and technology artifacts, such as IXP equipment and the passive infrastructure. Negotiations are continuous, and as players’ strategies and characteristics change, the relationships also change, strengthening or weakening an IXP’s equilibrium.

IXPs may have different deployments and pathways, depending on where they are built.

In fact, it is unlikely that one can just transfer an IXP from one country to another, given that actors will likely to be different in each territory, and will require, as a consequence, adaptations of other players, including in terms of design and governance when appropriate. Also, countries have different regulatory constraints. Because of that, definitions in which there is a locked understanding of an IXP such as in Fanou, Valera, et al. (2017), who state that an IXP is “where new participation is not rigorously constrained,” are clearly normative and not a generalizable concept as the Mexican IXP demonstrates.

157

“The Internet is only virtually stable” (Star & Bowker, 2010, 237). The present study reiterates that. It is not that IXPs are formed and then expected to be permanent. Incomplete translation processes can generate discontinuation provoked by a chain of actors or yet fragile policy results.

While associating a technological artifact with reducing the digital divide, strengthening sovereignty, and other pertinent outcomes may work as public narratives for actors seeking for support to infrastructure projects as IXPs, the level of technology determinism embedded in such political discourses, and its limitations, become clear as the politics of internet interconnection is unveiled.

The lack of competition in the Mexican market and the dominance of Telmex are some barriers for the IXP, but what the present analysis also highlights is that understanding power relations underlying the internet also requires shifting attention to the availability of passive infrastructure, optical fiber links and critical internet resources. This is when internet policy analysis gains substance, by making visible internet materiality and the role of the internet infrastructure itself.

In Chapter 2, shared networks emerged as responses to indigenous communities’ lack of connectivity, but the used interconnection solutions showed limitations and a high level of constraints to local networks’ autonomy. Despite the promises of the IXP in Mexico City to address digital divide issues, its long distance from Oaxaca and Chiapas, regions that pursue the lowest connectivity rates in the country, along with the low number of ISPs in the region to provide affordable fiber links to the capital, makes the first IXP in Mexico an unlikely solution

158

for the problems faced in those communities. There is in fact a lost link between this policy solution and the reality found there.

As an alternative, a regional IXP, with the goal of connecting different indigenous communities in that region to share local content directly through their own servers, could be a possible start to leverage local production as is already seen in San Martín Abasolo, Chiapas. For that to happen, communities could, for example, build direct point-to-point networks to a place where all the networks can afford to reach it with wireless lines. A pioneer in building the first and nowadays biggest IXPs in Europe, Klaus Landefeld, when asked to suggest a solution for that these networks, says that they could build an IXP at a very unlikely location, distant from big city centers, but where passive infrastructure such as radio towers is available and accessible to all the communities: “a natural point to traffic capacity exchange,” focusing on their own local content.

Following the availability of resources in the area, this possible interconnection arrangement would work independent of the internet for now, embedding communities’ values to the design and governance of such interconnection artifact with a local content orientation.

Nothing prevents, in the future, regional ISPs from becoming interested in interconnecting to that regional interconnection point, following the governance rules established by the communities.

As shared networks become more resourceful, they may opt to get access to autonomous system numbers in order to have access to the internet interconnecting to an IXP already built. At an IXP, it is possible to keep network autonomy, in internet terms, and establish peer relations with other networks of interest, such as CDNs of big content providers that dominate global

159

internet traffic. Whether this is enough for the communities considering local values and projections for the future is a question that they should answer.

As already stated in Chapter 1 and will be further discussed ahead, an IXP alone is hardly sufficient to provide full internet access to any peer because it depends on the networks connected to it and the routes that they make available. It also depends on the types of economic arrangements made between the entities connected, given that some tend to exercise their autonomy by not peering or doing businesses with potential competitors. Interconnection arrangements are the entrance to the global internet and present numerous constraints similar to ticket gates.

As the present chapter on the interconnection politics of the first IXP in Mexico elucidates, an IXP built to serve private ISP networks will tend to orbit around business-oriented issues. For it to connect to social goals, the inclusion of other voices in the governance of IXPs should be considered to consolidate them as public interconnection artifacts in a very public sense. Nevertheless, for new players such as indigenous communities’ networks, who become part of a regular IXP, while creating their own mechanisms to make non-neutral infrastructure serves their intents, the constraints that the IXP design imposes will always need to be taken into consideration.

In the next chapter, I present a more systemic view of the global IXP ecosystem, showing that the recent deployment of the Mexican IXP is an example of trends in the development of

IXPs in the global South as a whole. The assumption is that by looking symmetrically to IXP deployment in the global South and global North, the differences between the two and the state of internet infrastructure in the South will become more evident.

160

CHAPTER 4

THE SOCIETY OF INTERNET EXCHANGE POINTS: ASYMMETRIES BETWEEN THE GLOBAL NORTH AND THE GLOBAL SOUTH

The cyberspace domain is often described as a public good or a global commons, but these terms are an imperfect fit. A public good is one from which all can benefit and none should be excluded, and while this may describe some of the information protocols of the Internet, it does not describe the physical infrastructure, which is a scarce proprietary resource located within the boundaries of sovereign states and more like a “club good” available to some, but not all.

— Joseph S. Nye Jr., The Regime Complex for Managing Global Cyber Activities (p. 6)

This chapter aims to advance understanding of the internet interconnection ecosystem in the global North and the global South. It shows the politics embedded in the internet exchange points precursors, their roots in the United States and early dissemination to Europe. By examining the current IXP ecosystem based on merging four main IXP datasets available, Euro-IX, Packet

Clearing House, PeeringDB and Telegeography, this chapter indicates that our capacity to map

IXPs and attest that they are active is still limited. Despite that, data resultant confirms the unequal distribution of IXPs in the North and South, and critically approaches the latter development of IXPs in the global South.

161

1. The Origins of IXPs

The history of the internet in the United States is dependent on the deployment of network interconnection solutions. As it is generally known, starting in 1969, the U.S.

Department of Defense developed the ARPANET (Advanced Research Projects Agency

Network), a network administered by the Advanced Research Projects Agency (ARPA), which initially served as a research network, but was increasingly used for operational military purposes as well (Abbate, 1999; Roberts et al., 2011). Such different uses of the ARPANET impeded government efforts to generalize the level of security to military standards. A solution to that impasse was to divide the network, in 1983, moving the strictly military activities to

MILNET (Military Network), and keeping the research activities at the ARPANET (Abbate,

1999; Roberts et al., 2011). Two years later, in 1985, the National Science Foundation (NSF) created NSFNET (National Science Foundation Network), a less government-centered network which encompassed supercomputer centers, laboratories, regional and academic networks in the

United States (Abbate, 1999). Other government agencies were also developing their own networks, which would later interconnect.

Archived government agencies’ webpages show that under a high-level cabinet council chaired by the U.S. President, the National Science and Technology Council, there was a second- tier department, the Federal Networking Council (FNC), which hosted the Federal Engineering

Planning Group (FEPG) (NITRD, 1991, 1997). In 1989, members of the FEPG designed and engineered two interconnection points called Federal Internet eXchange (FIX) (MERIT, 1994)— one located in the University of Maryland, College Park, known as FIX East, and the other one

162

in Palo Alto, California, known as FIX West. The purpose of the FIX structures is explained in a document of 1995:

From 1986-1995, the National Science Foundation has provisioned a backbone network to support the activities of the academic and research community. The Dept. of Defense, NASA, and Dept. of Energy also provisioned networks to support specified groups of scientists. In order to avoid as much duplication of effort as possible, the various agencies decided to connect the communities that they served at two interconnection points (FIX). (NITRD, 1995, online)37

This excerpt implies that the interconnection points are technological artifacts to connect dissimilar governmental research networks to optimize their resources. Going further, FIXes were designed to connect government networks that are technically and politically administered differently: they allowed a Department of Defense network to connect to other academic networks under government administration, assuming types of governance autonomy among them.

In another document, the Michigan Educational, Research Information Triad, or MERIT, the company that contributed to build FIXes, explains that “With this new Internet infrastructure in place, the ARPANET was phased out by mid-1990” (MERIT, 1994, P. 8). In the context of the history of the internet, the discontinuation of the ARPANET can be considered the beginning of the migration of internet’s formal control from government to the private sector, FIXes’ or interconnection points’ deployment is intimately related to that.

Parallel to the shift from ARPA to NSF control and the connection of all main networks, a movement to amplify the number of nongovernment networks under the NSFNET was in place. However, even under NSF administration, there were government constraints that prevented further use of the network for commercial purposes (Abbate, 1999), and the private

37 A backbone network is a large network that connects many local networks (Denning & Martell, 2015, p. 222). 163

sector started to pressure for competition in the provision of backbone services. The impediment for the private networks to interconnect on the internet backbone administered by NSF was one of the important reasons for three new Internet Service Providers (ISPs) – PSINet, CERFNet and

Alternet – to form the nonprofit Commercial Internet Exchange (CIX) in July of 1991 (Abbate,

1999; Metz, 2001). With a meeting place to connect, traffic and routes could be exchanged directly among these providers and they could have access to the larger internet of that time

(Metz, 2001). Beyond the government restrictions to commercial traffic, another motivation for this new arrangement was that as a result of internet privatization in the early 1990s, there were charges and fees to interconnect through a private backbone that these ISPs wanted to avoid

(Abbate, 2010).

Once more, looking to the history of the internet from the standpoint of exchange points shows that their emergence is a technical necessity responding to a political one, that of including actors left outside the initial internet project, in the case of CIX, and connecting networks that wanted to retain their own control, as was the primary goal of FIXes. Whereas FIX formed an inter-net of government actors, CIX formed an inter-net of private actors.

Interestingly, the CIX arrangement added value to the network operators, who could now connect to other networks’ users through it, a fact that attracted other operators beyond the

United States’ borders: “With the commercial networks imposing no restrictions on the type of traffic they would carry [as the NSFNET imposed], the CIX became, in effect, a commercial version of the Internet, offering the same set of connections to a different clientele” (Abbate,

1999, p. 199).

164

As soon as the government allowed private network traffic, NSF invested further in the idea of interconnection points. It granted contracts to four companies to operate four new interconnection points that would connect the government networks and the commercial networks. These facilities were called Network Access Points (NAPs) and their primary locations were Chicago, Pennsauken NJ, San Francisco, and Washington DC (Metz, 2001). In

Chicago, for instance, it was operated by Ameritech, a company originated from AT&T divest in the 1980s, and that after merges and acquisitions, is again part of AT&T (AT&T, n.d.). These interconnection points set the conditions for network expansion and marked the full transition to internet privatization (Abbate, 2010; Abbate, 1999). Regarding design, NAPs were supported by switches—devices that facilitate receiving and forwarding traffic on the internet— and were founded on traffic and route exchanges that are still the basis of today’s IXPs.

According to an interview with George Strawn, a mathematician who was involved with the NSFNET deployment and was later a Chief Information Officer (CIO) at NSF, exchange points were expected to introduce competition in the early days of the internet, favoring the inclusion of new players. In a broader context, there were visible concerns at that time with the role of telecommunication carriers in the emerging data provision services. As the internet was built over existing telecommunications infrastructure, the Federal Communications Commission

(FCC) in the United States expressed concerns about barriers for entry in what was called “data market” in view of common telecommunications carriers’ initial advantages. The regulator created safeguards to “ensure that competing data providers had nondiscriminatory access to the underlying communications components of their service offerings” (Oxman, 1999, p. 10).

165

Historically, since the 1970s, in the United States, data services were distinguished from telecommunication services, and in the 1980s, the FCC ratified this understanding, creating two terms: “basic” and “enhanced” services. The latter was defined as a service “offered over common carrier transmission facilities” (Oxman, 1999, p. 10), which unlike the former, should be unregulated and included the brand-new internet.

These events at the government level not only opened the door for a new market; they also enabled a new digital sector noticeably regulated by business rules, and “code,” meaning the technology design itself, as per Lawrence Lessig’s types of regulation terms (Lessig, 2006).

These events are an important background to the study of IXPs considering that technologies are a result of numerous variables that surround them and influence their deployment process

(Abbate, 2010; Abbate, 1999; DeNardis, 2009; Dourish, 2015; Jackson et al., 2007). IXPs are even today regulated by code, standards, engineering norms understood as good practices, and the market.

1.1. New Commercial Arrangements in the Face of New Technology Design

Government sponsorship and non-law regulation values on the one hand, and private initiative and competition values on the other, were embedded in the initial design of interconnection points, and laid the groundwork for the origins of internet interconnection economics. The business model for the CIX arrangement was based on a membership fee paid by each network interconnected, while the traffic was defined to be free of charge (Abbate, 1999).

The reasons given for this option vary. Gerald W. Brock emphasizes the “mutual benefit” noticed by the participants of such arrangement (Brock, 1995). Janet Abbate, in turn, considers 166

that the computer network design and the unterritorial way of how it was understood by customers was key for such choice to avoid technical complications to manage and charge network traffic:

Unlike the telephone system, computer network customers were not charged on the basis of how far their packets traveled; indeed, they often did not even know the physical locations of the computers to which they sent packets, and they were even less likely to know which commercial network a computer was on. Under these circumstances, trying to impose charges for sending packets between networks would have caused great technical difficulties and, in all likelihood, would have outraged customers. (Abbate, 1999, pp. 198-199).

This lack of awareness about the route of one’s data is still true: no users know where their data is traversing or originating, and no singular networks know the exact route that data packets will follow in advance, as it depends on each network with which they have peering or transit relations and the decisions that each individual network router takes each time packets traverse them.

Yet regarding transit, it is worth noting that in the very beginning of IXPs, it was prohibited to occur at IXPs because customer-provider relationships were considered incompatible with the collaboration model initially embedded at IXPs. It was when IXPs started to be conceived as businesses at the end of the 1990s, and started to become more service- oriented, that different kinds of networks were permitted to be part of these interconnection points, offering a diverse range of services, including transit, web hosting, content distribution, etc. (Metz, 2001).

In 1999, William B. Norton, who worked at MERIT, and ended up being one of the founders of Equinix—a huge IXP player of the contemporary internet—wrote that “Scaling the

Internet requires not just trenches full of dark fiber, but an infrastructure that permits a rich set of interconnections among the networks that make up the Internet” (Norton, 1999, p. 1). He 167

defended the creation of “The Neutral Internet Business Exchange Model,” to reorient the market where ISPs and carriers were common IXP operators, creating possible conflicts of interest for potential competitors to interconnect and collaborate. Such facilities would allow for carriers to develop fiber arrangements and sell transit to other ISPs, which in turn would be there not only to buy available services, but also to peer with content providers and other ISPs, presumably giving economic advantages to all parties involved. This seeming neutrality is embedded in the design that underlies IXPs not governed by carriers or ISPs to this day, although governance models differentiate among IXPs, even if they work on the same “neutral” basis.

The creation of IXPs in a region at the first years of the beginning of the commercial internet represented primarily the possibility of keeping local traffic local; however, as occurs today, if the data requested in a region was hosted in another region (e.g. a website requested in

Europe that is hosted in the United States), IXPs would not necessarily prevent international traffic. In this regard, archive research shows early discussions on data caching as a solution for the high cost to access data in the U.S. As the minutes of a Working Group on Caching meeting in 1997 shows:

Currently more than 40 countries are exploring efforts to reduce traffic on the saturated links to the U.S. […] John Dyer provided some statistics on Internet usage in Europe illustrating the need for caching. More than 70% of the hits are for .COM addresses in the U.S. As a first step, the group moved parts of the .COM domain into a Europe-based cache. Early results of the establishment of that cache resulted in a drop to 40% of the requests for .COM going to the U.S. […] (Since the cache is a research project, the issues related copyrights are not addressed.) Future plans for the task force include adding another TLD [top level domain] like .ORG to the cache.” (NITRD, 1997b)

John Dyer is a technologist who has worked in education and research networking deployment in Europe since that time. Most importantly, discussions on the need for caching that

168

are very vivid today in the global South were already ongoing projects in Europe in the beginning of the commercial internet.

This brief historical narrative reveals technical, economic and political factors as interrelated in the formation of internet interconnection points, positioning these facilities as good vantage points to understand internet deployment and privatization, a topic of primary concern in contemporary internet governance scholarship.

2. A Study on Internet Exchange Points Datasets

The number of IXPs in the world is an important metric to help understand the state of the public internet interconnection ecosystem. However, arriving at this number is surprisingly complicated and varies depending on source and approach used. This difference can be a result of inaccuracy and/or different collection methods that characterize the current databases available (Klöti et al., 2016) and shows that counting and mapping IXPs are challenges per se. In this section, by merging the four main IXP databases available, which are Euro-IX, Packet

Clearing House (PCH), PeeringDB (public datasets), and Telegeography (private dataset), one of the goals is to present the number of registered IXPs as of February 2018. This is the second effort to merge IXPs datasets in the literature to the best of the author’s knowledge. The first, conducted by Klöti et al. (2016), focused on the three public databases and found an increase of

40.5% of the IXPs available when comparing to a singular source as PeeringDB.

The present dataset consolidation is original for including one more dataset, well known among the internet technical community and made available by the company Telegeography.

Following previous work, the present consolidation uses the IXPs’ name and city to match the 169

cases, and adopts a conservative approach to the matches, in that matches may not be realized due to lack of information, but false matches are less likely to happen. Unlike Klöti et al. (2016), the merging procedures were manually conducted, and formulas were used to highlight similarities and disparities in the fields of interest.

2.1. Limitations

Although IXPs’ websites were consulted to provide additional information in order to facilitate the merges, and consults pointed to the fact that there were inactive IXPs, such IXPs were not excluded from the datasets, as a systematic check of active and inactive IXPs was not a focus of the present research. In fact, this is a difficult answer to give with the currently available information. Authors already pointed to the limitations of IXP datasets in terms of freshness and accuracy as such data is mostly informed voluntarily and not kept up to date as consultations of

IXPs websites confirm (Chatzis et al., 2013; Klöti et al., 2016).

Thus, the database resultant of the merge is not a census of active IXPs, but a census of registered IXPs in public and private datasets, which aims to leverage the community knowledge of IXPs in terms of their total number, location and date of establishment. Answers about the actual activity of current IXPs is an important agenda for future research.

Data was either collected through the datasets’ websites or received through e-mail from a dataset administrator between February 6 and 8, 2018. For comparative studies, data should be ideally collected with the shortest period possible between datasets. As one of the sources was not retrieved, but sent by the administrator, and it was necessary to recollect data of one of the datasets afterwards, there is a difference of one to two days between the sources’ dates.

Nevertheless, the impact of this in the level of matches should not be expected to be significant, 170

as these datasets are relatively dynamic, but are not likely to suffer big changes in a very few days.

2.2. Considerations on Datasets

Euro-IX: Is short for European Internet Exchange Association, which is currently reformulating its IXP directory to become a more comprehensive data source synchronized with other datasets. On February 8, 2018 I retrieved 387 IXPs visible in their website (12.2% less than the 441 used in the 2016 study for being considered active). This is the smallest source in the present collection, retrieved in the very moment of Euro-IX transition to a new database, as indicated in the organization website, at the time of writing, that its new IXP directory includes

“some static data from the old Euro-IX website as of February 9th, 2018” (Euro-IX, n.d.).

Information collected for the present study from Euro-IX used to be provided by IXP operators, and affiliated members were expected to have more accurate registers than non-affiliated (Klöti et al., 2016).

PeeringDB: Is a popular dataset used by the networks to facilitate their peering arrangements. Data was collected through its application programming interface (API)38 on

February 7, 2018, when 643 IXPs were retrieved, from which 3 were excluded for being duplicated, resulting in 640 IXPs (33.3% more than the 480 used in the 2016 study for being considered active). Information in the IXP dataset is self-informed and can be used by

38 Available at https://peeringdb.com/api/ix. 171

autonomous systems to support peering decisions in conjunction with data from other

PeeringDB’s datasets on networks and facilities for peering.

PCH: Is short for Packet Clearing House, an international organization that works on topics related to IXPs and Domain Name System (DNS), and invests in keeping a comprehensive

IXP directory in which the majority of information is organized by PCH staff (70%), and the rest is gathered through the contribution of the internet community and IXP administrators (Klöti et al., 2016). PCH is the only dataset that was downloaded with IXP status (Active, Defunct,

Deprecated, Not an IXP, Planned, Senacty, Unknown), an information that may not be completely accurate (Chatzis et al., 2013; Klöti et al., 2016). For the present work, we used the

Active and Unknown statuses only for that database, which resulted in the retrieval of 626 IXPs on February 6, 2018, from which 2 were excluded for not having an IXP name and one for being duplicated, resulting in 623 IXPs (66.6% more than the 374 used in the 2016 study for being considered active). For the others dataset, without a status, all IXPs available were considered.

Telegeography: Is a market research and consulting company focused on telecommunications that keeps an IXP directory online whose functionalities provide limited access to download it. Through e-mail communication, the organization provided a list of 505

IXPs available in their database as of February 7, 2018, from which 3 were excluded for being considered duplicates, resulting in 502 IXPs. This database is important for the present study because along with PCH they are the only sources that provide dates of establishment for IXPs, although a limited number of them has this information as will be seen.

To facilitate the match analysis, the present consolidation adds a match factor (MF) to indicate in how many databases the IXPs are listed: MF 1 indicates that a given IXP is listed in

172

only one database while MF 4 indicates that the IXP is present in all four datasets. The main imprecisions of the datasets found in the merging process are listed as follows, explaining the use of the MF when pertinent:

• Duplication: Databases may have IXPs listed more than once. For instance,

PeeringDB lists both New York International Internet eXchange and NYIIX,

while, similarly, Telegeography lists Nap of the Americas and Nap Miami, but

both cases in New York and Miami are the same. In case a given IXP is listed in

one more database, in the source where it is duplicated, one register was merged,

receiving MF 2 as an indicator of its appearance in two datasets, and the other one

was excluded. Below are the seven duplication cases found. While some have the

same name, others :

Table 2. Duplication Cases in PCH, PeeringDB and Telegeography Datasets

PCH PeeringDB Telegeography Kept Deleted Kept Deleted Kept Deleted Equinix-NY Equinix New BBIX Tokyo Asia Smart IX Nap do Brasil Equinix São York IBX [BBIX Asia] Paulo IX NYIIX New York NAP of the NAP Miami International Americas Internet eXchange KRS-IX RED-IX IX Cuiabá MT IX Cuiabá MT

• Missing data: Name and city are the main fields to match IXPs at a first level. If

one of these fields is blank, that IXP will receive MF 1, for not being possible to

merge. For instance, Euro-IX lists CyrusOne as an IXP in Carrollton, and another

CyrusOne without a location. The first one was merged with an IXP with the

173

same name and city listed in PCH, receiving a designation of MF 2, while the

second received MF 1.

• Different registration pattern: While PCH, PeeringDB, and Telegeography

commonly list IXPs by their names, Euro-IX may list them by the name of the

organization that run IXPs instead. For example, Euro-IX lists NIC.br in São

Paulo, while PeeringDB lists it as IX.br (PTT.br) São Paulo, PCH lists it as Ponto

de Troca de Tráfego Metro São Paulo, and Telegeography lists the same as PTT

Metro Sao Paulo. For this kind of match to be realized, further website

consultation was conducted, as variations on the names lead formulas to indicate

no-matches, and manual analysis is definitive to consolidate the merge. In case

that an organization runs more than one IXP, it needs to have a city to be

matched, Otherwise, the organization register will not be merged and will receive

MF 1.

Yet on different patterns of registration, if an IXP has no city registered in a

dataset, but its name matches with another dataset, both will be merged and will

receive MF 2. For instance, in Euro-IX, the British Virgin Islands Internet

Exchange was registered without a city, indicating the country British Virgin

Islands, while in Telegeography, it was registered with a similar name, British

Virgin Islands IX, indicating the country Virgin Islands (U.K.) and the city Road

Town.

Finally, if an IXP has been registered with more than one city in the same field

(e.g. Dallas and Texas), that solo register will become two after the merge with a

174

dataset where each city corresponds to an IXP. This is how the number of IXPs in

the Euro-IX database increased from 387 in the beginning to 390 in the end of the

study. For instance, although INX-ZA, NAPAFRICA and Norwegian Internet

eXchange were listed only once at Euro-IX, as each indicated two cities in its

register, the three were merged two times each.

Additionally, Klöti et al. (2016) caution that IXPs may have different points of

presence and still be considered administratively just one entity. However, they

notice that the datasets are not consistent in keeping the same level of detail when

listing IXPs. As a result of their effort, 5.5% of the total registers in Euro-IX,

6.3% in PeeringDB and 7.1% in PCH datasets were reduced to 1.4%, 2.5%, and

2.7% registers in the datasets respectively when points of presence of the same

IXPs were merged among them. Thus, this provides the estimation of 4.1%, 3.8%

and 4.4% of inflation in the total number of IXPs in Euro-IX, PeeringDB and

PCH datasets, respectively, when such factor is taken into consideration.

• Lack of updates: In all databases, defunct IXPs or old IXPs that currently have a

different name may still be listed. For instance, Telegeography lists both Nap do

Brasil and Equinix São Paulo as different registers, but as contextual information

indicates, the latter replaced the former years ago, so this generates a duplication.

Not unusually, this case is registered differently by other databases: PCH and

Euro-IX keep its old name, while PeeringDB indicates both names as synonyms

as it has two name fields in the dataset (name, and name long). In the present

175

merge, Nap do Brasil/ Equinix São Paulo received an MF 4 for being in all

databases, and the duplicated case in Telegeography was excluded.

It is important to notice that while a match factor (MF) = 1 may be resultant of problematic facts as cited above, and be a terminated or defunct project, IXPs with that factor may exist and be active. In the same direction, although a MF > 1 points to more consistency, as it shows that an IXP is present in more than one database, it does not necessarily guarantee that a given IXP is active. Moreover, the analysis of the PCH database based on status and match factor shows that 9% of IXPs with active status in that dataset have MF 1. Also, while IXPs with unknown status are much more likely to have no matches—68% versus 9% of IXPs with active status have MF 1—a considerable amount of 32% of them present MF > 1.

Table 3. PCH Dataset per Status and Match Factor

Active Unknown 1 9% 68% 2 10% 14% 3 38% 14% 4 43% 4% Total 100% 100%

Base: 623 IXPs

Regarding the date of establishment, the development of IXPs worldwide shows imbalances from which the late formation of the first IXP in Mexico as examined in Chapter 3 is just one example. I conduct a systematic analysis of the late deployment of IXPs in the global

South based on data available in two of the IXPs databases in which there are dates of establishment: PCH and Telegeography. On that, Klöti et al. (2016) have already reported scarce information on IXPs establishment dates in the PeeringDB, PCH, Euro-IX datasets, while the 176

Telegeography dataset shared with the researcher by this company for the present dissertation study has such information for roughly 50% of the IXPs. Nevertheless, data available in the two datasets has not been consolidated yet, and this will be the goal here.

2.3. Results from the Merge Process

2.3.1. Comparison with previous research

For the purpose of comparison, I adapted a table created in Klöti et al. (2016) with authors’ study results of the consolidation of three datasets to allow a rough comparison with the present research. As our methods differ, and the datasets collected in 2014 and 2018 are also different in size and received unlike treatment before the merge, results were expected to be different. However, the comparison shows no highly disparate results.

In Table 4, the datasets are indicated in the first three columns. The values of intersection come next indicating the number of IXPs matches found between the datasets in the present study and in the 2016 study. The intersection percentage difference column helps to synthesizes that. Such indicator shows very low variation in the studies regarding the comparison of the three datasets altogether (1.1%) and of Euro-IX and PCH (3.1%). Conversely, it shows 21.2% more matches between Euro-IX and PeeringDB, and 37.4% fewer matches between PCH and

PeeringDB in the 2016 study.

177

The union of the datasets is the result of the intersection value resulted of all the

combinations possible among the datasets, plus the IXPs that were not possible to merge.39 As

expected from the larger datasets used for the present study, the union resultant in all

combinations is from 23.2% to 29.5% smaller in the previous study than in the present.

Table 4. Comparison of Intersection and Union in Euro-IX, PCH and PeeringDB Datasets Merge

Klöti et al., Klöti et Dataset 2019 2019 2016 Intersection al., 2016 Union Percentage Percentage Euro-IX PCH PeeringDB Intersection Intersection Difference Union Union Difference

   270 273 1.1% 876 673 23.2%   293 355 21.2% 737 566 23.2%   294 303 3.1% 719 512 28.8%   460 288 37.4% 803 566 29.5%

Table format adapted from Klöti (2016)40

Large union sizes are resultant of proportionally fewer matches in view of the datasets

size in the present study as observed in the Jaccard index depicted in Table 5 below. This index

shows how similar the datasets are among them, and is resultant of a division of the intersection

by the union (Klöti et al., 2016). It shows that while in the 2016 study the three datasets

presented 40.6% of similarity, in the present study they are more dissimilar, and this index

decreases to 30.8%. When read associated with the overlap found, there is an indication that the

reduction of similarity is better explained by variation in dataset sizes. The Overlap w/ the

39 As an example, the union of Euro-IX, PCH and PeeringDB is a result of the intersection of the three datasets (N=270), plus the intersection of Euro-IX and PeeringDB (N=23), Euro-IX and PCH (N=24), PeeringDB and PCH (N=190), and the IXPs not merged among them in Euro-IX (N=73), PeeringDB (N=157) and PCH (N=139), totalizing 876 IXPs. In Klöti (2016) there is no information about the calculus, and I am assuming that the logics behind it is the same. 40 I am grateful to Daniel Ribeiro and Deen Freelon for additional suggestions to improve Tables 4 and 5. 178

Smallest index also in Table 5 is resultant of the intersection divided by the smaller dataset, or, in other words, the extent to which the smallest dataset is contained into the larger. As showed below, it had a small variation, from 73.0% to 69.2%, when the merge of the three datasets was considered, and shows no huge differences considering other combinations in the two studies, ranging from 69.2% to 75.4% in the present research and 73% to 81% in the previous one.

Table 5. Comparison of Jaccard and Overlap Indexes in Euro-IX, PCH and PeeringDB Datasets Merge

Klöti et Klöti et al., Datasets 2019 2019 al., 2016 2016 Overlap w/ Overlap w/ Euro-IX PCH PeeringDB Jaccard Jaccard the Smallest the Smallest    30.8% 40.6% 69.2% 73.0%   39.8% 62.7% 75.1% 80.5%   40.9% 59.2% 75.4% 81.0%   57.3% 50.9% 73.8% 77.0%

Table format adapted from Klöti (2016)

Overall, this comparative profile of research projects with different approaches but with analogous intents points to a consistent picture in which available IXPs datasets are considerably different among them—even more in the present study where the datasets sizes have an important weight. Variations between the studies may be resultant from different research approaches as well as changes in the datasets themselves along four years that separate the collections. Initiatives to permanently consolidate the sources available for a better understanding of the current IXP scenario are welcome.

2.3.2. Synthesis of the Merge of Four IXPs Datasets

179

The merge of Euro-IX, PeeringDB, PCH and Telegeography datasets generated a list of

914 IXPs, as shown in Table 6. This corresponds to 42.8% more IXPs than the largest dataset used in the merge process, PeeringDB (N=640)—a result consistent with previous work findings, in which the compilation of three datasets focused on listing active IXPs resulted in information for 40.2% more IXPs than in PeeringDB (Klöti et al., 2016, p. 3). While the present result cannot lead one to affirm that there are 914 active IXPs in the world, due to datasets limitations already pointed out, even considering a possible inflation in the data due to siblings IXPs around 4% according to previous study, this number is yet impressive and much higher than a singular dataset indicates.

In joining four datasets, the intersection resulted in 222 IXPs, with a level of similarity among datasets being noticeably low (24.3%). Even when considering the merge conducted with three and two datasets, it is impressing that although all datasets are trying to portray the same reality regarding IXPs distributed worldwide, their level of similarity is still considerably low— ranging between 28.0% and 44.3% in the merge of three datasets, and between 38.7% and 60.4% in the merge of two.

The Overlap w/ the Smallest and the Overlap w/ the Largest indexes in each merge further show how different in size the datasets are. For instance, in the main merge, encompassing the four datasets, the intersection corresponds to 56.9% of the smallest dataset,

Euro-IX, and 34.7% of the largest, PeeringDB, meaning that more that 56% of Euro-IX is contained within PeeringDB and less than 35% of the latter is contained into Euro-IX. At another end, Table 6 also interestingly shows that 85.7% of Telegeography dataset is contained into the

PeeringDB when only both are compared.

180

Finally, considering the metrics altogether, the most similar datasets are PCH and

PeeringDB, not only in terms of content, with 57.3% of similarity, but also in terms of size, as

more than 70% of each dataset are contained into the other—73.8% of PCH is contained into

PeeringDB and 71.9% of the latter is contained into the former.

Table 6. Synthesis of Euro-IX, PCH, PeeringDB and Telegeography Datasets Merge

Datasets Overlap Overlap Jaccard Intersection Union w/ the w/ the Euro-IX PCH PeeringDB Telegeography Index N=390 N=623 N=640 N=502 Smallest Largest     222 914 24.3% 56.9% 34.7%     270 876 30.8% 69.2% 42.2%     233 832 28.0% 59.7% 37.4%     233 793 29.4% 59.7% 36.4%     375 846 44.3% 74.7% 58.6%   294 719 40.9% 75.4% 47.2%   293 737 39.8% 75.1% 45.8%   249 643 38.7% 63.8% 49.6%   460 803 57.3% 73.8% 71.9%   404 721 56.0% 80.5% 64.8%   430 712 60.4% 85.7% 67.2%

Table format adapted from Klöti (2016) with data from the present research.

Table 7 below depicts the IXPs distribution according to their match factor. It shows that

public sources have IXPs exclusive to their databases, or MF 1, in 17.4% (Euro-IX), 17.7%

(PeeringDB) and 19.4% (PCH) of the cases. Telegeography, the only private source, is the one

with the lowest proportion of exclusive IXPs, 7.6% of its dataset. Summed up, this generate a

merged dataset with 37.2% of IXPs with MF 1, which could not be merged.

181

Table 7. Match Factor per Euro-IX, PCH, PeeringDB and Telegeography Datasets

Merged Dataset Euro-IX PCH PeeringDB Telegeography

N % N % N % N % N % 1 340 37.2% 68 17.4% 121 19.4% 113 17.7% 38 7.6% 2 129 14.1% 30 7.7% 68 10.9% 93 14.5% 67 13.3% 3 223 24.4% 70 17.9% 212 34.0% 212 33.1% 175 34.9% 4 222 24.3% 222 56.9% 222 35.6% 222 34.7% 222 44.2% Total 914 100.0% 390 100.0% 623 100.0% 640 100.0% 502 100.0%

A continuation phase that I envision for this project is to do a collaborative check on the

914 IXPs resultant of the study, by building an API in which the IXP and internet communities altogether contribute. With a simple interface, volunteers would select their region, country or city and “certify” that a given IXP is active. This could be called IXP Activity Status API, which I consider a more reliable approach than relying on status informed in IXP datasets that, against efforts in contrary, may yet contain stale data. The number of “certifications” got through the

API can provide an additional reliability indicator of activity status—the more certifications the more reliable is the status informed. Furthermore, such API can serve as a continuous update service when synchronized with current static datasets, providing a necessary and currently inexistent input for a better understanding of IXP scenario worldwide.

2.3.3. IXPs Unequal Distribution Worldwide

Numbers show an unequal distribution of IXPs between the global North and Global

South, in a reason of 60% versus 40%, as shown in Table 8.

182

Table 8. IXPs Distribution between the global North and global South

N % Global North 548 60.0% Global South 366 40.0% Total 914 100.0%

When looking at the ten top countries with the higher number of IXPs in both global

North and South in Table 9, the United States assume the top of the list. Interestingly in the global South, all the BRICS (Brazil, Russia, India, China and South Africa) are listed.

Table 9. Top 10 Countries in IXPs Distribution between the global North and global South

Global North N % Global South N % 1 United States 200 36% Russian Federation 49 13% 2 Germany 30 5% Brazil 37 10% 3 Poland 27 5% Argentina 29 8% 4 France 26 5% China 24 7% 5 Canada 24 4% India 22 6% 6 Australia 23 4% Indonesia 20 5% 7 Japan 22 4% Ukraine 17 5% 8 United Kingdom 19 3% Philippines 10 3% 9 Netherlands 16 3% Bulgariaa 9 2% 10 Sweden 15 3% South Africa 7 2% a The World Bank lists Bulgaria as an upper middle income country, alike Brazil, Russia and others.

As the numbers in Table 9 are generally higher than the usual numbers that appear in a single dataset—for instance, the U.S. has 128 IXPs (88 active) and Russia has 30 IXPs (27 active) according to the PCH dataset, —it is useful to note the match factor of IXPs listed within countries.

First, as seen in Table 10, the proportion of IXPs with MF 1 is similar between countries in the global North and South, between 36.5% in the first case and 38.3% in the second.

183

Table 10. Distribution of IXPs with MF 1 IN the global North and global South

MF 1 Total N % Global North 548 200 36.5% Global South 366 140 38.3% Total 914 340

When considering the countries that have more IXPs with MF 1, both U.S. and Russia lead the list once more, as shown in Table 11. For instance, 45% of the IXPs with MF 1 within the global North are based in the U.S, indicating that further research to know how many are active will be welcome. Interestingly, checking the website of the organization that run the public IXPs in Brazil, there are by the time of this writing 31 active IXPs—fewer than the total of 37 depicted in Table 9, but more numerous than the 29 IXPs with MF > 1 found in the present study (NIC.br, n.d.). Nevertheless, United States in the global North, and Argentina, Brazil,

China and Russia in the global South are leaders in number of IXPs.

Table 11. Distribution of IXPs with MF 1 in the global North and global South per Country

Global North Global South Country N % Country N % United States 90 45.0% Russian Federation 20 14.3% Poland 18 9.0% Indonesia 14 10.0% Canada 11 5.5% China 11 7.9% France 9 4.5% India 10 7.1% United Kingdom 8 4.0% Ukraine 9 6.4% Japan 8 4.0% Brazil 8 5.7% Germany 7 3.5% Philippines 6 4.3% Netherlands 7 3.5% Ecuador 4 2.9% Chile 5 2.5% Bulgaria 4 2.9% Switzerland 4 2.0% Pakistan 3 2.1% Other 33 16.5% Other 51 36.4% Total 200 100.0% Total 140 100.0%

184

As important as the countries that have more IXPs, are the countries that may have and have not IXPs. In the merged database, 148 countries present at least one IXP listed, which are shown in Table 12.

Table 12. Distribution of Countries with Listed IXPs per Continent and per global North and global South

Global North Japan 22 Singapore 6 Taiwan, China 6 Korea, Rep. 6 United Arab Emirates 3 Asia Bahrain 2 Qatar 2 Cyprus 1 Israel 1 Hong Kong SAR, China 1 Saudi Arabia 1 Asia Total 51 Germany 30 Poland 27 France 26 United Kingdom 19 Netherlands 16 Sweden 15 Italy 11 Spain 8 Switzerland 8 Norway 7 Ireland 6 Europe Czech Republic 6 Denmark 5 Austria 5 Lithuania 5 Finland 5 Estonia 3 Slovak Republic 3 Latvia 3 Hungary 2 Luxembourg 2 Belgium 2 185

Portugal 2 Croatia 1 Liechtenstein 1 Malta 1 Slovenia 1 Iceland 1 Greece 1 Europe Total 222 United States 200 Canada 24 Panama 3 North America British Virgin Islands 1 Trinidad and Tobago 1 Barbados 1 Puerto Rico 1 North America Total 231 Oceania Australia 23 New Zealand 10 Oceania Total 33 Chile 9 Sint Maarten (Dutch South America part) 1 Curaçao 1 South America Total 11 Global North Total 548 Global South South Africa 7 Tanzania 4 Nigeria 3 Kenya 3 Swaziland 2 Congo, Dem. Rep. 2 Namibia 2 Cameroon 2 Egypt, Arab Rep. 2 Tunisia 2 Africa Gambia, The 2 Morocco 2 Angola 2 Zimbabwe 2 Sudan 1 Rwanda 1 Ghana 1 Lesotho 1 Sierra Leone 1 Liberia 1 186

Burundi 1 Zambia 1 Benin 1 Botswana 1 Senegal 1 Mauritius 1 Burkina Faso 1 Congo, Rep. 1 Gabon 1 Mozambique 1 Togo 1 Côte d'Ivoire 1 Uganda 1 Djibouti 1 Réunion 1 Madagascar 1 Malawi 1 Africa Total 60 China 24 India 22 Indonesia 20 Philippines 10 Thailand 5 Cambodia 4 Iran, Islamic Rep. 4 Pakistan 3 Vietnam 3 Bangladesh 2 Nepal 2 Georgia 2 Asia Lebanon 2 Iraq 2 Kyrgyz Republic 2 Timor-Leste 1 Sri Lanka 1 Mongolia 1 Bhutan 1 Malaysia 1 Armenia 1 Uzbekistan 1 Palestine 1 Kazakhstan 1 Lao PDR 1 Asia Total 117 Russian Federation 49 Europe Ukraine 17

187

Bulgaria 9 Romania 4 Turkey 3 Moldova 3 Albania 2 Serbia 1 Macedonia, FYR 1 Belarus 1 Montenegro 1 Kosovo 1 Europe Total 92 Mexico 2 Jamaica 1 Belize 1 Nicaragua 1 Cuba 1 Honduras 1 North America Dominica 1 Costa Rica 1 Martinique 1 Dominican Republic 1 St. Lucia 1 Grenada 1 Haiti 1 North America Total 14 Vanuatu 2 Oceania Papua New Guinea 1 Oceania Total 3 Brazil 37 Argentina 29 Ecuador 6 South America Colombia 4 Bolivia 2 Peru 1 Paraguay 1 South America Total 80 Global South Total 366 Grand Total 914

As one more indicative that these numbers may contain inactive or defunct IXPs, while the list presents 60 IXPs located in Africa, The African IXP Association (AFIX), which has put an effort to keep an up-to-date list of IXPs in the region online, indicates that there are 44 active African 188

IXPs instead, distributed among 32 countries (AFIX, n.d.). The phenomenon of how IXPs die and become inactive is an interesting topic for further study. As seen in Chapter 3, local dynamics with incumbents may put IXPs in fragile equilibrium.

Comparing such list with the 252 countries resultant from the merge of United Nations

(UN) and World Bank (WB) list of economies, it results in 104 countries with no listed IXPs— which can also be interpreted as probably no history of IXPs in such territories. Interestingly, such countries are proportionally distributed between the global North and South as shown in

Table 13, however, as seen in Table 14, global North territories with no listed IXPs are composed mainly by islands, while the global South territories encompass numerous African,

Asian and Latin American countries.

Table 13. Distribution of Countries with no IXPs between the global North and global South

Total Countries Countries with no (UN and WB) IXPs listed N % N % Global North 91 36.1% 39 37.5% Global South 161 63.9% 65 62.5% Total 252 100.0% 104 100.0%

Table 14. Distribution of Countries with no Listed IXPs per Continent and per global North and global South

Global North Africa Seychelles Heard Island and McDonald Antarctica Islands Brunei Darussalam Christmas Island Cocos (Keeling) Islands Asia Kuwait Macao SAR, China Oman Åland Islands Europe Andorra Channel Islands 189

Faroe Islands Gibraltar Guernsey Holy See Isle of Man Jersey Monaco San Marino Sark Svalbard and Jan Mayen Islands Antigua and Barbuda Aruba Bahamas, The Bermuda Cayman Islands North America/ Greenland The Caribbean Saint Pierre and Miquelon St. Kitts and Nevis St. Martin (French part) Turks and Caicos Islands Virgin Islands (U.S.) French Polynesia Guam New Caledonia Oceania Norfolk Island Northern Mariana Islands Palau South America Uruguay Global South Algeria Cabo Verde Central African Republic Chad Comoros Equatorial Guinea Eritrea Ethiopia Africa Guinea Guinea-Bissau Libya Mali Mauritania Mayotte Niger Saint Helena São Tomé and Principe

190

Somalia South Sudan Western Sahara Bouvet Island French Southern Territories Antarctica South Georgia and the South Sandwich Islands Afghanistan Azerbaijan British Indian Ocean Territory Jordan Korea, Dem. People's Rep. Maldives Asia Myanmar Syrian Arab Republic Tajikistan Turkmenistan West Bank and Gaza Yemen, Rep. Europe Bosnia and Herzegovina Anguilla El Salvador North America/ Guadeloupe The Caribbean/ Guatemala Central America Montserrat Saint Barthélemy St. Vincent and the Grenadines Global South (cont.) American Samoa Cook Islands Fiji Kiribati Marshall Islands Micronesia, Fed. Sts. Nauru Niue Oceania Pitcairn Samoa Solomon Islands Tokelau Tonga Tuvalu United States Minor Outlying Islands Wallis and Futuna Islands Bonaire, Sint Eustatius and Saba South America Falkland Islands (Malvinas) 191

French Guiana Guyana Suriname Venezuela, RB

Regarding this reality of countries with no IXPs, in which ISPs are dependent on incumbents only and Tier-1 operators to transmit international traffic, organizations as Internet

Society has promoted actions, training, and provided funding to change the current scenario in order to increase the quality and decrease the price of the internet in such countries. According to

Michuki Mwangi, from Internet Society in Kenya, while there are already successful IXPs in

Africa, from which he mentions Kenya Exchange Point and South Africa Exchange Point as examples, the history of the Africa internet, marked by monopolies, has prevented some IXPs projects to take off—a phenomenon that deserves further research in which the countries above can be good loci for field work.

2.3.4. IXPs’ Late Development in the Global South

To examine the development of IXPs in the global North comparatively with the global

South, IXPs’ establishment date information provided by PCH and/or Telegeography datasets is the main variable analyzed in this section. It is important to note, though, that not all registers of these datasets contain dates, and even when they are complete, there are possible mismatches among the two sources. As a result, out of 914 IXPs in the merged dataset, the analysis is based on a subtotal of these IXPs, as will be further explained below.

PCH and Telegeography are the only sources to provide establishment date information; the first provides it for 536 IXPs (86.0% of its 623 registers), and the second for 279 (55.5% of 192

its 502 registers). As seen in Table 15, PCH alone contributed dates for 298 IXPs. Telegeography alone, in turn, provided dates for 41 IXPs. Both datasets shared date information for 238 IXPs.

Table 15. Date Information Source

Date Information N PCH 298 PCH & Telegeography 238 Telegeography 41 No IXP in PCH and Teleg 193 Blank field (no data) 144 Total 914

To further show how datasets differ among themselves, in 77 of the 238 cases in which date information was available in both sources there was date mismatch. The analysis of these 77 cases is shown in Figure 11 below. Most of the mismatches are caused by one (36 IXPs) or two years of difference (13 IXPs) in the informed dates among the two datasets. As for the analysis proposed here, two years are not significant to change the course of analysis: with these 49 cases,

30 from the global North and 19 from the global South, I adopted the criteria to use the oldest date in the mismatch, assuming that it is less likely that an IXP would be registered with a date before it exists, than that it would be registered afterwards with a delay. The remaining 28, with mismatches from 3 to 18 years of difference, were discarded for the present analysis.

193

Dates Difference in Years - PCH and Telegeography 40 35 30 25 20

# of of # Cases 15 10 5 0 0 2 4 6 8 10 12 14 16 18 20 Difference in Years

Figure 11. Dates difference in years between PCH and Telegeography

Additionally, to build a historical timeline in years, I disregarded 10 IXPs built in 2018, as data was collected in early February 2018, preventing a consistent estimation of a trend for that whole year. With these procedures, the final analysis incorporates 539 IXPs. Figure 12 shows that in the early internet development, most IXPs were built in the global North, with a great disparity with the global South at least until 2005-2009, when the gap starts to be considerably reduced. Interestingly, it seems the IXP birth trend starts to change in 2010-2014 both in the North and in the South, although in the latter the down curve is less accentuated, indicating that the IXP birth rate may start to be higher in the global South than in the North.

194

IXPs Buit per Period - 1993-2017

79 70 76 64 55 54 56 35 35

7 8 19 -3 0 -8 1993-1994 1995-1999 2000-2004 2005-2009 2010-2014 2015-2017 -7 -35 -47

Global North Global South Difference

Figure 12. IXPs Built per Period – 1993-2017

Looking at the IXPs that were built in the first period, the London Internet Exchange

(LINX) and the Equinix-Palo Alto stand out. Between 1995 and 1999, in the global South, the

CABASE IXP Buenos Aires, NAP do Brasil, NAP Chile and NAP Colombia in Latin America,

Johannesburg Internet Exchange in South Africa, Hong Kong Internet Exchange, and MSK-IX

Moscow were built. In the North, the Amsterdam Internet Exchange (AMS-IX) and the

Deutscher Commercial Internet Exchange (DE-CIX) in Frankfurt stand out in that period.

Interestingly, with a higher IXP rate growth in the 2000s, it is only in this period that other IXPs in Africa started to be built, including the Cairo Internet Exchange, Kenya Internet Exchange

Point, Mozambique Internet Exchange, Rwanda Internet exchange, Tanzania Internet Exchange, and Uganda Internet Exchange. It was also between 2000 and 2004 that Brazil, through a multistakeholder body, started to build a public IXP ecosystem under a discourse of sovereignty, as will be seen later. Nevertheless, the development of the global South is by no means uniform, and responds strongly to local dynamics as seen in Chapter 3. Figure 13 shows IXPs’

195

deployment differences in Africa and South America, which started in 1996 with the

Johannesburg Internet Exchange, in South Africa. The dotted lines are trending lines, showing more IXPs being built in South America than in Africa during the same period of time.

IXPs Built in Africa and South America - 1993-2017 10

8

6

4

2

0

-2

Africa South America Linear (Africa) Linear (South America)

Figure 13. IXP Built in Africa and South America – 1993-2017

During the Internet Governance Forum 2016 in Guadalajara Mexico, I talked with Jean-

Robert Hountomey about the development of IXPs in the Africa. Hountomey is a member of the

African Internet Exchange Task Force, and former head of Internet Society Chapter in Togo and founder of AfricaCert, an organization focused on cybersecurity issues in the region. He shed light on the problems that the deployment of IXPs in Africa in the 2000s aimed to solve, as well as on important contextual factors that have negatively influenced the development of an IXP in

Togo, which happened only in 2017. Aspects associated with the lack of local infrastructure, regulators’ approach that favor telco incumbents, and institutional instability are some of the issues that arise from this interview and set the ground for a continuous dependency on services from the global North.

196

The Motivation for the Emergence of IXPs in the 2000s and Today’s Challenges

There were three issues. One was in terms of the local traffic, the [use of] international link and [the] coming back. The second was latency and delay issues. Around 2005, most of countries were using satellite connectivity. With satellite, you have at least five hundred milliseconds for one connection. You have to go international and come back to reach the network of the ISP in the country and that takes you to one thousand milliseconds, what was not acceptable for services. Also, the international bandwidth was very expensive (and still is). So, it was interesting to build inside countries internet exchange points. Now, we are in the phase 2 of the project, where between neighbor countries, sometimes, even two countries being under fiber, sea cable, etc., with no agreement between them, they still have international traffic (...) [For example,] Benin has an IXP, Togo is building it. Now, the traffic between Togo and Benin goes to Italia Telecom… Then, [the approach is] to discuss with the countries and see if they can put a link together to avoid that. To go further on that idea, we thought about having a kind of regional carrier, like a country acting as a hub for other countries, (…) so that the local traffic within that region stay local.

Possible Decision Drivers for a Country to Become a Hub

It is based on the connectivity of that country, the wire ability, intensity of electric, and power connection... There are still countries that still have power issues... Some countries do not have several landing points to fiber cable [either].

The Late Development of the First Internet Exchange Point in Togo

When I was the chair of ISOC, there was a debate in that many [interested parts] broke up. ISOC took the challenge and started to talk with the ISPs, and put them together (…). We went to see the regulator and speak to him what an internet exchange point is so that they can allow us to build that interconnectivity without paying those fees [that used to be charged in interconnection arrangements]. We have been talking about that for five years now [2016]. [It takes long] because of change of government. We need the involvement of the local regulator in the internet exchange point also to make sure that we don’t get into trouble.

Since our conversation, the African Union has endorsed the development of communication infrastructure and the “African Internet Exchange System,” with support of the

European Union-Africa Infrastructure Trust Fund and the Government of Luxembourg (Africa

Union, n.d.). As government leaders recognize IXPs a priority to reduce broadband costs in the region, Jean-Robert Hountomey informs that IXPs deployment has evolved positively. 197

Due to the late IXP development in the global South, even with higher rates of development in the 2000s, only in 2015 has the increasing gap in relation to the global North started to be reversed as seen in Figure 14, where the accumulated IXP growth is depicted. In regard to the numbers below, although they represent the IXPs with date information in PCH and

Telegeography only, the proportion of IXPs in the global North and South that come from this data—57.5% and 42.5% respectively—is consistent with the 60.0% and 40.0% shown in Table 8 for the whole merged dataset, allowing the Figure 14 to be a good longitudinal portrayal of the

IXP ecosystem.

IXP Ecosystem Development - 1993-2017 310 275 229 196 175 132 99 62 43 7 0 8 1993-1994 1995-1999 2000-2004 2005-2009 2010-2014 2015-2017 -7 -54 -81 -89 -97 -100

Global North Global South Difference

Figure 14. IXP Ecosystem Development – 1993-2017

For a world map representation of this deployment, see Figures 15, 16 and 17 below.

198

Figure 15. Map: IXP Ecosystem Development – 1993-2000

199

Figure 16. Map: IXP Ecosystem Development – 1993-2010

200

Figure 17. Map: IXP Ecosystem Development – 1993-2017

201

3. Final Considerations

It is thought-provoking that in times of increasing control over data, the society of IXPs is still marked by rich yet inaccurate databases, and inaccessible information complicated to assess.

Nonetheless, by joining different sources and providing a longitudinal analysis of IXP creation, this chapter contributes to a better understanding of these artifacts, and their unequal distribution and deployment with regard to global North and global South countries. With 914 IXPs resulting from the consolidation of four different sources, this chapter suggests that the number around

600 IXPs generally used to refer to the universe of IXPs worldwide is underestimated. To what extent these are active IXPs and not sibling IXPs, a second phase to certify IXPs’ statuses with the internet technical community is suggested.

With a symmetric analysis of the global North and the global South IXP ecosystems, the asymmetries that the data expose open the possibility of investigating further the interconnection dynamics and the flow of information that occur within and between the two regions, as we shall see in Chapter 5. Specifically, the fact that some of the biggest IXPs in terms of traffic worldwide are based in affluent areas in the North, including the AMS-IX in Amsterdam, the

DE-CIX in Frankfurt and LINX in London—all built in the 1990s—raises the question about how attributes historically associated with IXP definition, as keeping the local traffic local, are less valued in scenarios where some IXPs become globalized and highly attractive to international traffic. Parallel to that, IXPs in the South that were created under the narrative of sovereignty, as were the more than 30 Brazilian IXPs—starting in the 2000s—also raise the question of how they position themselves in the face of these globalized dynamics. 202

CHAPTER 5

INTERNET INFRASTRUCTURE INTERDEPENDENCIES AND THE CONCEALMENT OF THE GLOBAL SOUTH

The process that creates the global as the dominant position in unequal exchanges is the same one that produces the local as the dominated, and therefore hierarchically inferior, position. In fact, we live as much in a world of globalizations as we live in a world of localizations.

—Boaventura de Sousa Santos, Globalizations (p. 396)

This chapter proposes a new way of looking at internet infrastructure by analyzing symmetrically internet interconnection dynamics in leading IXPs in both the global North and the global South.

With a comparative approach, it shows how giant IXPs in the North are nourished by uneven infrastructure conditions in the South, shaping contemporary communication with infrastructure interdependencies between the North and the South, and a one-way flow of information from the first to the latter. It argues that big content providers become key actors in interconnection politics, as they have highly demanded content worldwide but are physically present in only some facilities, following their own private policies. With that, symbiotic relations between giant

IXPs and big content providers are formed. The dynamics revealed are an example of continuous patterns of inequality, and the coloniality of power distinctive of the South.

1. Introduction

According to Boaventura de Sousa Santos, globalization is defined in terms of “unequal exchanges” in which a local artifact (or a condition or an entity) gains prominence and global 203

status, differentiating itself from other local artifacts across national borders. For the author,

“there is no originally global condition; what we call globalization is always the successful globalization of a particular localism” (Santos, 2006, p. 396). With that, he calls attention to the political aspects that the term globalization implies and the consequent “disappearance of the

South,” in that processes of economic integration and interdependences weaken the problematization of the South and its distinctions from the North in view of the global economy.

For him:

Since, contrary to this discourse, the inequalities between the North and the South have dramatically increased in the past three decades, this fallacy seems to have no other objective than to trivialize the negative, exclusionary consequences of neoliberal globalization by denying them analytical centrality (Santos, 2006, p. 395, emphasis added).

More than a decade has passed since the author wrote this, and yet it seems a good basis from which to approach IXPs with a global and systemic view. With a South perspective, this chapter asks: is the internet interconnection infrastructure different when viewed from the vantage point of the global South? It aims to shed light on commercial dynamics and infrastructure conditions that have contributed to IXPs in the North becoming leading IXPs worldwide. By analyzing the global North and South symmetrically, through the materiality of the internet interconnection infrastructure, this chapter endeavors to avoid “trivializ[ing] the negative” (Santos, 2006, p. 395), making evident some patterns of inequality that arise from the analysis.

204

2. The Free Flow of Information Paradigm

In the field of Global Communication, a key element of globalization that has been critically discussed is the “free flow” of information. On the one hand, the concept has a general positive meaning within the United States, where it may be associated with “a regulatory mechanism” in the market of media (Hepp, 2015, p. 62), or yet with the protection of both freedom of speech and access to knowledge from government interference in line with the First

Amendment (Mcdonald, 2004). On the other hand, when applied to global communication, “free flow” gains characteristics of a boundary object, whose “interpretive flexibility” (Star, 2010) and understanding as a one-way flow is relevant here. I will keep its quotation marks due to this fact.

In a transnational arena, the “free flow” paradigm emerged strongly in the midst of the

Cold War representing the Western countries’, and precisely, the United States’ position on the development of international media (Thussu, 2018). For Daya Kishan Thussu, “The ‘free flow’ doctrine was essentially a part of the liberal, free-market discourse that championed the rights of media proprietors to sell wherever and whatever they wished” (Thussu, 2018, p. 41). And given where the media resources at that time were already concentrated, that position would bring more gains to Western countries, including their governments and media owners, the author concludes.

In regard to the United States, according to Soenke Zehle, instituting the “free flow” as a

U.S. foreign policy principle was a response to the U.S. media market pressure on its government. They wanted to change the scenario during World War II in which European countries had prominence in transnational communication. The United Kingdom, along with

France, was preponderant in media news production within their colonies, while the UK also had 205

prominence on sea cable infrastructure (Zehle, 2012). Interestingly, the course of actions within the post-war United Nations system, especially in the United Nations Education, Scientific and

Cultural Organization (UNESCO), show that the “free flow of ideas” doctrine was ratified as a

UNESCO principle in its constitution from 1945 by the initial group of twenty member states such as Brazil, China, Egypt, India, South Africa, along with the United Kingdom, the United

States, France and other Western nations (UNESCO, 1945), indicating the paradigm resonating beyond the countries considered the ones who would benefit most from it.

The critical turn to the “free flow” paradigm would arise some years later, when in the

1950s, the Soviet Union and allies became members of UNESCO and positioned against it, on the grounds that it was seen as an interference in state sovereignty (Zehle, 2012). Such critics had affinities with a larger group of Asian and African countries that in the 1960s formed the Non-

Alignment Movement (NAM). The NAM was a result of broader political efforts to counter the

Cold War bipolarism, and set the ground to strengthen common interests among countries with a colonial history, aggregating keen political forces around its ideals (Rajak, 2014). According to

Tukumbi Lumumba-Kasongo, “The establishment of the Non-alignment Movement in 1961 was intended to begin the process of actualizing solidarity and cooperation among all nation-states, which were willing to join a block of interests called the Global South” (Lumumba-Kasongo,

2015, p. 11). These emergent actors later formed the group of 77 countries (G77) within

UNESCO (Lumumba-Kasongo, 2015), helping to shape the “New World Information and

Communication Order” (NWICO) agenda at a moment when independence movements and the consequences of colonialism were in evidence, and a broader proposal led by NAM countries for

206

a “New International Economic Order” (NIEO) was already a guideline within the United

Nations (Oppermann, 2018; Zehle, 2012).

The New World Information and Communication Order (NWICO) agenda was shaped by a general dissatisfaction with the disparities characteristic of global communication—among which the critics of the “free flow” of information were part. UNESCO became the main forum for NWICO debates between 1972-1985, and the promotion of these issues was one of the reasons for the United States and the United Kingdom to have left the organization in the 1984 and 1985, respectively (Zehle, 2012). Among the strong points of the agenda were cultural imperialism through media as well as controversies regarding satellite broadcasting by the

United States that, due to its transnational reach, was understood by others as an outrage to national communication regulations (Zehle, 2012). And while the critique of cultural imperialism at that time was related to many different media productions, including films, the concentration of international news agencies in France, United Kingdom and the United States producing 80% of the world’s news emerged as a key example of it (UNESCO, 1980, p. III-9).

The UNESCO report “Many voices, One World,” also known as the MacBride report— named after the head of the International Commission for the Study of Communication

Problems, responsible for the research—is an important hallmark for the NWICO agenda.

Regarding the “free flow” of information following the considerations of critical countries, the report ratifies that “the theory of ‘free flow’ is invalidated by the overwhelming preponderance of information circulated from a small number of industrialized countries into the huge areas of the developing world” (UNESCO, 1980, p. III-6). Such imbalance is known as “one-way flow,” as opposed to a “two-way flow” of information. Behind these discussions is thus the control of

207

technology and content resources in the hands of a few countries. Independent of the media and the time of this discussion, the access to its infrastructure is key to understanding the distribution of content. Also, the characteristics and affordances of the infrastructure have a role in such dynamics—as the satellite controversy clearly shows.

In internet times, it has been a general understanding that the internet was created based on a bottom-up philosophy that would allow others to increment, improve and transform its technologies. Lawrence Lessig called this “Open-Evolution” (Lessig, 1999), and Jonathan

Zittrain called it “generative Internet” (Zittrain, 2008). These values—openness and generativity—can be associated with the ideas of unpredictable deployment, and the capacity to create innovation and disruption. In line with Western-liberal perspectives, these are not incompatible with proprietary ownership as both authors agree (e.g. a proprietary operating system that allows for the creation of new platforms based on it is generative). I bring these references to contextualize possible new readings of “free flow” of information in the internet era, in which these imagined bottom-up characteristics have been associated with the ideal of an open and free internet. For some, such an internet ideal is a rhetoric in defense of human rights in the digital context; for others it is embedded on liberal principles in that low barriers to entry in the internet economy would allow new market choices to arise. In this case, resultant inequalities are seeing with naturality.41 Nevertheless, the possibility of materializing such ideal has been increasingly challenged by facts, such as market concentrations that in the content level are

41 An example of that is an old Lessig’s explanation that “The market guarantees open entry, which simply means it guarantees that everyone may enter the competition. But to guarantee equal entry is not to guarantee equal results. Winners are to be chosen by the market’s choice” (Lessig, 1999, p. 1418-1419). 208

marked by the U.S. Google, Amazon, Facebook and Apple, or GAFA domination (Moore &

Tambini, 2018).

I propose to problematize the “free flow” of information from an IXP standpoint by examining IXPs situated in the North and in the South symmetrically. The assumption is that this approach will allow us to reflect on the formation of localisms, in Boaventura de Sousa Santos’ terms from the perspective of the formation of the internet as a global communication medium where infrastructure imbalances lead certain locals to become global as it will be revealed.

3. Methods: Code Ethnography

In the sections that follow, I assume the role of a code ethnographer to examine the work of routers through the language that they use, the Border Gateway Protocol (BGP) when circulating data from one point to another on the internet. If BGP is the lingua franca for internet networks to communicate, do the network routers that are connected to an IXP in the North and an IXP in the South say different things when situated in one node or another?42

To answer this question, I analyze BGP data from two different IXPs: DE-CIX Frankfurt and IX.br São Paulo. I denominate this analysis code ethnography to call attention to where I am situated with respect to this language that is the BGP protocol and the population that is around

42 In Appendices A and B, I make an effort to introduce the basics of some communication technologies and protocols with a key role in the information circulation infrastructure of the internet, BGP and switches and routers communication included (Appendix B). The goals are to make visible how their technical attributes are embedded in social, political and economic relations—being sociotechnical above all, and to lay the groundwork for the analyses in Chapters 5 and 6, specially for whom do not have an engineering background like me. 209

it—including IXP switches, looking glasses, through which routing information may be seen, network engineers, routers, route servers, etc.

In view of my hypothesis regarding asymmetries between global South and global North countries in internet infrastructure and possible interdependencies, DE-CIX staffers were important to suggest to me to explore BGP data looking at the networks’ distribution as a proxy to the data flows from the IXPs, which they emphasized would be hard to get access to due to data private ownership and companies’ competitive issues associated with the disclosure of such data. I further investigated the existent possibilities with NIC.br staffers, the organization that runs IX.br and after having the viability check with these experts, Tiago Gonçalves—then a network engineer at the IX.br and at the time of this writing, at the Amsterdam Internet

Exchange—was the one who accepted to process the BGP data for further analysis.

To have access to the data, PCH was key. As a network, it is a regular participant of numerous IXPs in the world that collects daily data shots of what traverses its routers and makes it available to the internet community in the organization’s webpage.43 PCH BGP raw data allowed us to have access to DE-CIX routing information, not easily accessible through DE-CIX looking glass. In fact, although DE-CIX has a public application programming interface (API)44 that allows for consulting data individually, it does not facilitate bulk collection. And while the

IX.br in São Paulo was accessible to the research request for data, DE-CIX Frankfurt was not.

Thus, for routers communication signifiers to become manageable to the present research— which is in search of the types of internet infrastructure interdependencies visible when IXPs in

43 Available at: https://www.pch.net/resources/Raw_Routing_Data/ 44 Available at: https://lg.de-cix.net/ 210

the global North and the global South are examined symmetrically—, some steps were necessary.

Similar to other sites of ethnographic study, code ethnography is also dependent on inside informants who can guide the researcher through this field, which while it is non-territorial, is as material as the one that characterizes the indigenous network infrastructure investigated previously, as will be shown.

Below I provide a visual representation of a prefix announcement on BGP as a way to both materialize and explain what kind of data is the input for this chapter.45 Figure 18 contains information collected at an IXP route server, and is presented below as it is seen from the standpoint of an IXP’s member router. Data analyzed is based on the networks connected to the

IXPs (the autonomous system numbers highlighted with a red square in the Figure 18) and the origin networks announced (the autonomous system number in the most right). In the following paragraphs, the image is further explained.

Figure 18. A Border Gateway Protocol Script. Source: Built on PCH Routing Data

Starting on the left, one can see the prefix advertised 1.0.6.0/24, which is a block of 256

IP addresses relative to hosts that are reachable at this IXP. On the right of the prefix there are

45 For an introduction about routers’ communication, se Appendix B. 211

four BGP announces to this prefix, one per line. The first column indicates the router IP addresses correspondent to the autonomous systems that are IXP members, and whose numbers

(ASNs) are framed in red, and who are responsible for announcing that prefix for the IXP’s peers. Although there are four announcements, they are being done by just two autonomous systems connected to the IXP: AS3257 and AS6939, as seen in the first column. The first is announcing a unique path to reach that prefix, whereas the second is announcing three paths.

These routes are technically called Autonomous System Path (AS_PATH), and in the example above, all of them have four networks or hops to be followed from the IXP until they arrive to the AS56203, which is the owner of the prefix announced, or the origin network for that announcement. For example, the first path traverses the AS3257 – AS4826 – AS38803 until arrival to the origin network destination, the AS56203 (with a yellow square, in the most right).

Regarding the other three routes, although they present exactly the same hops, they are supposed to have different attributes—not visible in this image—that make them different paths. An example is that any autonomous system in the path may be connected to the neighbor in the left via more than one router for redundancy purposes, and this case characterizes two different paths. In terms of significance for data circulation, the more paths, the more alternatives for data packets to arrive to their destinations.

The path selected in this example has a “>” in front of it. The reasons for this selection are privately defined by network operators set up in the router policies. The Request for

Comments that define the current BGP version 4 informs seven path attributes (Rekhter et al.,

2006), whereas Cisco, the main router vendor in the market, presents thirteen rules to select the best path (Cisco, 2016). Guidelines to Internet Service Providers indicate that some of the main

212

attributes used to influence the selection are: WEIGHT, present in some vendors routers only and where the highest number is given to a preferred path; LOCAL_PREF (Local Preference), where the operator informs the highest number to the path that should be preferred; AS_PATH

(Autonomous System Path), where the shortest path should be selected; MED (Multi-Exit

Discriminator), where a lower number is defined to select the best path when there is more than one exit going to the same autonomous system, as shown by the case exemplified in the Figure

15 with three similar routes; and finally, preference for the route received first, which means the oldest one is also recommended (Cisco, 2016; DE-CIX, n.d.-b; Florentino, 2018). As an example of a rationale to set the values for the LOCAL_PREF attributes, according to IXP guidelines for its members, selection should prioritize first, routes internals to the autonomous system, then clients’ routes—given that clients are supposed to pay for the provider traffic—, followed by peering routes, commonly obtained at IXPs, and finally transit routes, generally considered more expensive than peer-to-peer arrangements (DE-CIX, n.d.-b). Behind this rationale is the goal of optimizing costs with the best quality of service. These decisions are under the control of autonomous systems; internet users do not know where their data traverses.

Yet regarding Figure 18, it is possible to suppose economic relations between the autonomous systems in the AS_PATH and to understand better how packets go from one point to another when traversing an IXP. Recapitulating, in the image, there are two autonomous systems directly connected to the IXP advertising the prefix 1.0.6.0/24 at this IXP: the AS3257 and the AS6939. They advertise that prefix because both are providers of the AS4826, which is a provider of AS38803 which, in turn, is a provider of AS56203, the origin network. In other words, it is because AS56203 is a client of AS38803 that the latter announces the prefix of the

213

former, and the chain continues until the autonomous systems that are connected to the IXP announce the prefix of the AS56203 to the IXP’s members. These relations are ultimately the assets shared in peering agreements as a form of collaboration among Internet Service Providers.

Before closing this explanation of the internet traffic design whose significance will be further clarified with the results of the present research in the next pages, it is important to remark on its governance. The so-called IP addresses, as well as autonomous system numbers, key to BGP communication, are considered “critical Internet resources,” unique identifiers under levels of centralized organizations’ governance (Laura DeNardis, 2014). As there are public interest implications that come from the control of such resources by the Internet Assigned

Numbers Authority (IANA), Regional Internet Registries (RIRs), and Local Internet Registries

(LIRs),46 they are good examples of the interrelation between internet governance areas and public policy issues deeply problematized in the literature (Mueller, 2010; Raymond &

DeNardis, 2015). Contrasting this scenario with the BGP governance, which likewise presents prominent levels of public interest involved, BGP control is not in the hands of centralized structures. Instead, it is under a distributed and decentralized administration of autonomous systems’ operators guided by their own private policies that directly impact the internet data flow. As possible governance centralization brings with it questions of power, legitimacy and authority, existing collaborations to leverage BGP routing security with hierarchical mechanisms have faced challenges and generated skepticism (Kuerbis & Mueller, 2017).47 It is amid the decentralized BGP design and the autonomy that network operators have to forward traffic, that

46 For explanations on these institutions, see Appendix B. 47 Resistance to some security solutions has also come from the fact that they require changes in both the routers and the BGP protocol, among other adaptations (Siddiqui et al., 2015). 214

internet exchange points become attractive nodes in the paths available, bringing overlooked centralization effects, as discussed in Chapter 1 and will be shown as follows.

4. Limitations

As already briefly explained, access to data that is privately owned is commonly difficult.

The alternative to accessing DE-CIX data resorting to the PCH routing repository generates some methodological considerations, nonetheless.

The routes that PCH’s devices receive depend on which other participants it is connected to. As the organization has an open connection policy and does not apply any filter to what its routers receive, the present research assumes that PCH’s repository has as many routes as a single participant at any IXP with a similar profile could have. However, some networks may not have an open connection policy and may not interconnect with PCH, which means that the view from a PCH’s router, as from any other IXP participant’s router, is always a partial view. On the other hand, data from IX.br SP comes from its API and represents the view of a route server, a device that mediates the relations among all participants that establish multilateral peering. It is a different source of data, although it also provides a partial view, as there are participants that make only bilateral agreements and do not participate with the route server. Nevertheless, as DE-

CIX is a leading IXP in terms of traffic, if its data came from its route servers and not from one participant as PCH, the considerable differences that the research found in relation to the IX.br would tend to be even bigger, with more unbalances favoring the IXP from the global North, but without changing the analysis.

215

Another important consideration is where networks reachable at both IXPs come from, if from the global North or global South. To investigate that, the autonomous system numbers

(ASNs) of connected and announced networks were consulted at Whois APIs to identify in which country such numbers are allocated. Thus, the present research uses the country where the

ASN is registered to identify where the network is from. A limitation to that is that there may be companies that have ASNs from different countries. For instance, in this study, a network may be considered from a country in Latin America even if the company that owns the ASN is from the United States. This is a plausible example, as big corporations may acquire ASNs from multiple places to increase their chances to be prioritized as a route in local markets, or to have access to blocks of Internet Protocol addresses already scarce in their original regions in the global North. In this case, as in the previous example, the differences between North and South may be bigger than the present research found out, and because of that, would not change the direction of the analysis.

Nevertheless, although the limitation regarding the ASNs’ precise localizations exists, it is worth mentioning that in Brazil, at the time of this writing, there are seventeen ASNs allocated in other regions that have access to IPs delegated to Brazil, but keep their previous ASNs in their announcements. For instance, AT&T, China Telecom, Highwinds, and Microsoft are some of them. Such situations, that favor the analysis, are difficult to foresee in terms of the frequency with which they occur, as they depend on the Local Internet Registries policies and the autonomous system operators’ private decisions.

216

5. Considerations on DE-CIX and IX.br

5.1. DE-CIX

Strategically located in the middle of the European continent, as shown in Figure 19, the

Deutscher Commercial Internet Exchange (DE-CIX) is the biggest IXP in the world in terms of traffic. Its location in Frankfurt, as noted in Figure 16, and its establishment date in 1995, are interestingly related to the consolidation of the city as the European financial center, where the

European Monetary Institute was established in 1994, followed by the European Central Bank that took its place in 1998 at the same site.

Figure 19. Frankfurt Geographic Location. Source: Reproduction from Google Maps.

In an interview with the author, Klaus Landefeld, an internet pioneer in Europe based in

Frankfurt and participant of the formation of various IXPs in the region, explained that as fiber optics networks were necessary for the financial industry to develop, Frankfurt was one the first cities where the private sector invested in building a substantial fiber network in the region. At

217

that point, in the beginning of the commercial internet traffic, one of the infrastructure challenges was that content was based in the United States and network exchange was also made there, where some exchange points were already in use, facilitating traffic exchange of national and international networks. As traffic in Europe increased substantially, undesirable costs and latency were important drivers to mobilize Internet Service Providers (ISPs) to join efforts and create local exchange points.

The fiber made available by the Metropolitan Fiber System (MFS) in Frankfurt attracted

ISPs to the region. This American company started to offer their infrastructure services for a price much more attractive than that practiced by the telco incumbent. According to Landefeld:

“The cost for international connectivity was lower in Frankfurt than anywhere else. It was the only place where you could actually get alternative capacity other than from Deutsche Telekom.”

As European backbones were being built, their convergence to Frankfurt followed that fiber networks were available and carriers and ISPs were being established there. That added positive value for Frankfurt, where the private sector also started to build data centers gradually supporting the city to become an important internet node in Germany and later in Europe.

At the same period, ISPs were forming internet exchange points in other important cities in the region, including London (1994), Brussels, Paris (1995), Amsterdam (1997), among others. Submarine cables were also being built from the United States to Europe and among

European countries as well, establishing the infrastructure that would influence the flow of information in the following days. See Figure 20.

218

Figure 20. Fiber Optics Network in Western Europe core in 2001. Source: Reproduction from Rutherford et al. (2004).

DE-CIX was formed as a not-for-profit organization based on the mutual interest of three

Internet Service Providers that decided to connect their fibers to a switch under some sort of association. Since the 2000s, however, DE-CIX is a well-structured for-profit organization—now a third part to facilitate the physical and logical interconnection for the realization of peering collaborations among hundreds of ISPs in a commercial way. Such a move, according to Harald

Summa, DE-CIX Chief Executive Officer in personal communication with the author, was motivated by the limitations to growth as a not-for profit. In the same direction, Klaus Landefeld explains that it is very different to be an ordinary IXP versus a leading IXP due to the investment in hardware and software that is necessary to keep up with the high traffic that a successful node will attract:

Our infrastructure needs to support the exchange of diverse flows without ever blocking any transfer. Designing a distributed high capacity infrastructure requires you to buy very selected hardware [and] large scale switches (…). And then you need software to

219

program the switches to build the paths dynamically, manage the full capacity. You need very good software for that, and we have worked with some companies to develop that.

At the time of this writing, DE-CIX has nine chief officers, fourteen people allocated in

Business Development, four people in Research and Development, six people in Service,

Network and Infrastructure Engineering, and a number of other distributed in Finance,

Marketing, Customer Service, Information Technology Security, to name a few.

The governance of such structure is under eco - Association of the Internet Industry, which represents more than a thousand information technology and internet companies mainly from Europe and the United States. eco’s board has four executives and its staff comprises more than seventy people, some of whom work at DE-CIX as well. Complementing structure, DE-CIX has a Customer Advisory Board with five members, one that eco’s board allocates, and another four elected by DE-CIX. Beyond Klaus Landefeld, who is also part of eco’s board, the Customer

Advisory Board has representatives from the major German telco, Deutsche Telekom, LinkedIn, regio[.NET] and GoDaddy.

By the time of this writing, DE-CIX Frankfurt is only one out of the company’s 17 locations, some under DE-CIX International division, a clear example of IXPs globalization phenomenon. In Germany, they also have a presence in Hamburg, Munich, Dusseldorf and

Berlin. In the global North, they are also located in Marseille (France), Palermo (Italy), Lisbon

(Portugal), Madrid (Spain), and New York and Dallas (U.S.). In the global South, they have a presence in Chennai, Delhi, Kolkata and Mumbai (India), Moscow (Russia), Istanbul (Turkey), and Dubai (United Arab Emirates).

220

5.2.IX.br

IX.br (former known as Ponto de Troca de Tráfico—PTT) is a major project under the

Brazilian Internet Steering Committee (CGI.br) administration—a national governance institution with a substantive role for both local and global reasons. First, the instability of

Brazilian political life that engenders persistent policy discontinuity and institutional changes

(Frey, 2000) underscores the significance of this long-lasting organization, launched in 1995.

Second, CGI.br is considered a worldwide pioneering multistakeholder institution given that it was formed three years before ICANN.

A multistakeholder committee with a national scope, CGI.br is composed of representatives from government, the private sector, NGOs, technical and scientific communities. It is a non-regulatory body responsible for establishing “strategic directives related to the use and development of the Internet in Brazil” (Brasil, 2003). Previous work has categorized CGI’s actions into four pillars: promotion of multistakeholder dialogues, deployment of infrastructure projects, production of internet-related information and indicators, and promotion of capacity building (Glaser & Canabarro, 2016). The committee sends representatives to national and international forums (Brasil, 2003), especially those dedicated to internet technical and governance issues, such as ICANN and IETF,48 and parallel to that can provide specialized knowledge to subsidize the action of the Executive, Legislative and Judiciary by invitation.

48 In parallel, ANATEL, the state telecommunication regulatory agency, is the official Brazilian representative at the International Telecommunication Union (ITU). 221

To execute its various tasks, CGI.br has an executive office, the Brazilian Center for

Information and Coordination of dot-BR (NIC.br), a civil, private and nonprofit legal entity divided into at least six sub-areas. Among them, there are the Registro.br which administers and operates the domain name system acting as both registry and registrar,49 and the IX.br project, responsible for developing and implementing internet interconnection policies via the operation of national IXPs.

The IX.br project, under CGI.br governance and NIC.br direct administration, is considered the largest public IXP ecosystem worldwide (Brito et al., 2016). Unlike other experiences—where IXPs may be mostly administered by companies (e.g. United States), not- for-profit organizations (Europe in general and other regions), and to a lesser extent by governments, (e.g. India)—, in Brazil, IXP administration by a multistakeholder organization implies a “public good” meaning to the IXP ecosystem (Afonso, 2004), according to Carlos

Afonso, one of the first CGI.br counselors and IX.br project idealizers who continues:

In 2002, FAPESP50 unilaterally resolved to convert the IXP—a non-profit service that, like the ccTLD itself, should be considered a public good—in a business operation, selling it to a company in Miami (Terremark). Therefore, the main point of data exchange in Brazil started to be operated by a U.S. company. Currently this business is being reviewed due to a pressure from CGI.br, which has a national IXP strategy to be implemented soon (Afonso, 2004, own translation).

In fact, in 2004 the PTT São Paulo, nowadays IX.br São Paulo, was built under a national sovereignty and public interest narrative, and until today, it is possible to hear that the IX.br is a

“project to improve the Brazilian Internet, and not a business,” as stated by Tiago Gonçalves in

49 Registry is the organization that sets the rules about the domain names under a certain country code top-level domain (ccTLD)—.br, in the case of Brazil. Registrar is the organization that sells the domain names to the public. 50 FAPESP stands for Fundação de Amparo à Pesquisa do Estado de São Paulo (São Paulo Research Foundation), historically the first organization responsible for internet governance issues within the country. 222

an interview with the author. For example, it does not follow a common market practice, which is to establish a Service Level Agreement (SLA) with its participants—known as the compromise between the provider and its clients in terms of services, deadline for technical maintenance, etc. Additionally, until 2017, the IX.br would not charge any participant for its interconnections, as such service was financed with the registrar activities of the .br domain names. By the time of writing, the IX.br pricing is much lower than commercial IXPs. For instance, for a port of 10 Gigabits per second, the participant will pay less than USD 180 (R$

690) per month. For a comparison, in Mexico, for the same port it is charged $ 2,400, using U.S. parameters.

This sense of public good is anchored on the CGI.br governance model, which supposes a balance of stakeholders’ interests in the decision-making processes. Behind it, there is the assumption that multistakeholder models “have the potential to promote better decisions through broader inputs” (Almeida et al., 2015, p. 75), as stated by the CGI.br former coordinator Virgilio

Almeida, Carlos Afonso, CGI.br former member, and Demi Getchko, NIC.br President and internet expert member of CGI.br. Nevertheless, “public good” can be differently understood from the perspective of the private sector versus civil society representatives: “the views of the third sector [NGOs] are diametrically opposed to those of the telecommunication representatives,” Virgilio Almeida stated in an interview with the author. Because of that, disputes are constant, and the resultant national interconnection policy is a result of that.

223

6. Comparing DE-CIX and IX.br BGP Data

A popular metric associated with internet exchange points is the number of networks connected to them. The bigger this number, the more attractive it becomes to newcomers, as an excellent example of network externalities and the centralizing attributes of IXPs within the distributed internet infrastructure. Figure 21 shows how the IX.br São Paulo is the leading IXP worldwide in this regard, including when compared with the biggest IXP in terms of traffic in

Europe, DE-CIX Frankfurt included.

Networks Connected to IX.br SP and DE-CIX Fra

IX.br 1,720

DE-CIX 879

Figure 21. Networks Connected to IX.br SP and DE-CIX Fra. Source: IXPs websites, March 24, 2019.

These numbers are very dynamic and point to a growing scenario. Compared with data from February 2018 (1,512 and 798, respectively), IX.br experienced a grow of 13.8% and DE-

CIX 10.2% during the previous year.51

51 Julio Sirota notes that the number of IX.br connected networks is difficult to be kept up to date, as there is a certain turnover due to the characteristics of the Brazilian networks, most of them of small size. He estimates a 224

When the current scenario is seen from a global North vis-à-vis South lens, it is possible to notice some significant differences in the profile of the networks that connect to each node.52

Because DE-CIX is located in the North and IX.br in the South, it was expected that the networks connected to them would come mostly from their own regions, what in fact happens, as shown in Figure 22. However, the DE-CIX level of attractiveness to networks from the South is much higher (23.2% or 204 networks) than the IX.br level of attractiveness to networks from the

North (3.3% or 57 networks). As DE-CIX is a commercial IXP, such results are in line with the company’s aggressive marketing built on “Where networks meet”—a slogan that purposefully alludes to an internet without borders that positions this IXP as a global internet node.

Networks Connected to DE-CIX Fra and IX.br SP per Global North and Global South

96.7% 76.8%

23.2% 3.3%

DE-CIX IX.br

Global North Global South

Figure 22. Networks Connected to DE-CIX Fra and IX.br SP. Source: IXPs websites, March 24, 2019.

possible 10-15% turnover, but there is not an official source of information other than the IX.br website that could be used for the present analysis. 52 The analyses that follow are based on 1,706 (out of 1,720) and 878 (out of 879) networks connected to IX.br and DE-CIX, respectively, whose ASNs registration countries were identified in Whois and Cymrus databases. 225

Taking a look at the countries that these IXPs attract or their level of internationalization, the differences between these nodes become even more clear. Tables 16 and 17 show DE-CIX

Fra connected networks per country within the global North South. Germany represents 29.6% of the networks connected, followed by the U.S. with 9.7%. From the global South, Russia is the main country to be represented with 7.5% of the total networks connected to DE-CIX, followed by Bulgaria and Ukraine, also in Europe. Brazil is the next on the list, representing 1.9%, with 17 networks. Interestingly, a year before, in February 2018, there was only one Brazilian network connected to DE-CIX Fra, showing that this country represents 19.8% of this IXP’s connected networks grow during this period of time.

Table 16. DE-CIX Fra Connected Networks per Country from the Global North

Global North Continent/ Country N % Europe 537 61.2% Germany 260 29.6% United Kingdom 42 4.8% France 38 4.3% Netherlands 30 3.4% Austria 27 3.1% Italy 21 2.4% Switzerland 20 2.3% Poland 19 2.2% European Union 18 2.1% Sweden 8 0.9% Hungary 7 0.8% Luxembourg 7 0.8% Belgium 4 0.5% Czech Republic 4 0.5% Denmark 4 0.5% Ireland 4 0.5% Norway 4 0.5% Lithuania 3 0.3% Finland 3 0.3% 226

Gibraltar 2 0.2% Spain 2 0.2% Slovenia 2 0.2% Latvia 2 0.2% Estonia 1 0.1% Slovak Republic 1 0.1% Greece 1 0.1% Malta 1 0.1% Croatia 1 0.1% Portugal 1 0.1% North America 89 10.1% United States 85 9.7% Canada 4 0.5% Asia 45 5.1% Singapore 7 0.8% Hong Kong SAR, China 7 0.8% Cyprus 6 0.7% Korea, Rep. 5 0.6% Israel 4 0.5% Saudi Arabia 4 0.5% United Arab Emirates 3 0.3% Bahrain 2 0.2% Japan 2 0.2% Kuwait 2 0.2% Oman 2 0.2% Qatar 1 0.1% Oceania 3 0.3% Australia 3 0.3% Total Global North 674 76.8% Total Global North and South 878 100.0%

Table 17. DE-CIX Fra Connected Networks per Country from the Global South

Global South Continent/ Country N % Europe 127 14.5% Russian Federation 66 7.5% Bulgaria 21 2.4% Ukraine 18 2.1%

227

Romania 9 1.0% Turkey 3 0.3% Serbia 3 0.3% Moldova 3 0.3% Belarus 2 0.2% Macedonia, FYR 1 0.1% Bosnia and Herzegovina 1 0.1% Asia 44 5.0% China 6 0.7% Indonesia 5 0.6% Thailand 4 0.5% Vietnam 3 0.3% Malaysia 3 0.3% Iraq 3 0.3% India 2 0.2% Sri Lanka 2 0.2% Philippines 2 0.2% Armenia 2 0.2% Georgia 2 0.2% Azerbaijan 2 0.2% Pakistan 2 0.2% Mongolia 2 0.2% West Bank and Gaza 1 0.1% Iran, Islamic Rep. 1 0.1% Cambodia 1 0.1% Jordan 1 0.1% South America 18 2.1% Brazil 17 1.9% Colombia 1 0.1% Africa 15 1.7% South Africa 8 0.9% Namibia 2 0.2% Egypt, Arab Rep. 2 0.2% Mauritius 2 0.2% Angola 1 0.1% Total Global South 204 23.2% Total Global North and South 878 100.0%

228

DE-CIX international performance is even more impressive when compared with the distribution of connected networks to IX.br. Table 18 demonstrates that IX.br is primarily a national node, where 96.1% of its networks are from Brazil. Importantly, the flow of networks from the North connecting to the biggest IXP from the global South is very low, and even lower is the flow of other global South countries to this node, showing that the DE-CIX attractiveness to neighbor countries’ networks is also higher when compared to the IX.br. Figures 23-26 synthesize in maps DE-CIX and IX.br different national and international performances.

Table 18. IX.br Connected Networks per Country from the Global North and Global South

Global North Continent/ Country N % North America 39 2.3% Canada 1 0.1% United States 38 2.2% Europe 16 0.9% Austria 2 0.1% Czech Republic 1 0.1% European Union 3 0.2% Germany 2 0.1% Hungary 1 0.1% Italy 1 0.1% Netherlands 2 0.1% Spain 1 0.1% Sweden 1 0.1% United Kingdom 2 0.1% South America 2 0.1% Uruguay 2 0.1% Total Global North 57 3.3% Global South South America 1645 96.4% Argentina 2 0.1% Brazil 1639 96.1% Colombia 1 0.1% Paraguay 3 0.2%

229

Asia 2 0.1% China 2 0.1% Africa 1 0.1% Angola 1 0.1% Europe 1 0.1% Russian Federation 1 0.1% Total Global South 1649 96.7% Total Global North and South 1706 100.0%

Brazil is a country of continental dimensions, with 8.5 million km. In comparison,

Europe has 10.8 million km. In such a context, Germany is more comparable to São Paulo, as the first has approximately 357,000 Km and the latter 248,000 Km.53 Based on that, Figure 25 depicts from where in Brazil the networks come that connect to IX.br SP. To do that, similar to the global data regarding where a given ASN is allocated, this information for the ASNs acquired in Brazil is provided by Registro.br as a registrar. Looking at the results, while more than one third of the networks connected to it are from São Paulo (35.4% or 600)—the most populated state, and also the financial center of the country—, it is impressive that most networks originate from other states (64.6% or 1095), when one considers the existence of many IXPs throughout the country built with a regional development perspective.54

Comparatively, Figure 26 depicts closer to where the networks come from that connected at DE-CIX Frankfurt. As already stated, networks from Germany account for 29.6% (or 260),

53 Populations differ substantially though, as the Brazilian population estimated for 2019 is 210 million people, and the European population as of 2016 is 741.4 million people. In turn, São Paulo has a population of 44 million people and Germany of 82 million people. 54 This analysis is based on 1,695 cases, as 25 out of 1,720 networks connected at IX.br could not have their locations identified at Cymrus, Registro.br or Whois datasets. 230

which, in context, is a similar number to that found at IX.br SP and the networks from São Paulo connected there. Interestingly, Germany has approximately 30 IXPs in its territory similar to

Brazil. I will analyze these numbers considering the concentration that they represent in local and global terms in the following sections. As the data shown so far elucidate, while DE-CIX Fra has become a centralizing point in the internet interconnection infrastructure in an international context, so does IX.br SP in a national context.

Importantly, to be connected to an IXP requires considerable investments from autonomous system operators, and needs to be financially advantageous in comparison with other forms of interconnection as the purchase of transit from Tier-1 networks—and it historically is. Beyond that, the distance of a network to the exchange point is also important, as the transport to reach an IXP data center and interconnect there is also part of the costs of IXP interconnection. In the history of IXPs, such nodes have been created where the networks in need of leveraging their connectivity are geographically based. What the location analysis of IXPs’ connected networks shows, though, is that other elements have been significant to interconnection dynamics involving IXPs, and those have local and global dimensions.

231

Figure 23. Map: IX.br Connected Networks per Region and Brazilian State

232

Figure 24. Map: DE-CIX Connected Networks

233

Figure 25. Map: IX.br Connected Networks per Brazilian States 234

Figure 26. Map: DE-CIX Connected Networks per European Countries

235

6.1. Thinking Locally: The Role of IXPs in the Interconnection Strategies of Tier-3 ISPs

Tier-1 businesses have a paradoxical relationship with internet exchange points. On the one hand, IXPs work as their strongest competitors as evidenced by the move from a hierarchical to a flatter model of the internet shown in the literature. On the other hand, Tier-1 companies also benefit from interconnecting at IXPs, as such facilities concentrate potential Tier-1 clients at a unique point, facilitating the selling and purchase of transit services. Once potential clients and providers’ networks are interconnected to the same facility, it is only necessary to pay the data center and the IXP for a cross-connection between the two networks to establish the interconnection.

Tier-2 networks, in turn, are more resistant to interconnect to IXPs. As they commonly own considerable infrastructure resources, telco incumbents are less prone to invest in collaboration with competitors, seeking to profit from the advantage that they have: extensive private infrastructure accumulated along the years. Sharing such actives are seen as unfair and counter-productive.

Tier-3 ISPs are the actors with fewer apparent conflicts with IXPs, and the most interested in interconnecting at such facilities as their networks have limited reach, and the more access to others’ infrastructure and the routing information they have through peering agreements, the more competitive advantages they acquire by increasing connectivity, and reducing costs. This in part explains the success of IX.br SP.

Regional providers, or small providers, as the Agência Nacional de Telecomunicações

(ANATEL) (Brazilian Telecommunications Regulator Agency) calls them–technically Tier-3

236

ISPs—are focused on bringing the internet to very specific regions of the country, in which big telco corporations offer unsatisfactory or no services at all, as informed by the president of the

Associação Brasileira de Provedores de Internet e Telecomunicações (Brazilian Association of

Internet and Telecommunication Providers) (ABRINT)’s Council, Breno Vale (Evans, 2016).

Unlike Mexico, where these conflicts were also seen but the licensed small providers are yet few,

Brazil has a dynamic market pushed by regional providers. For instance, in the second semester of 2018, ANATEL had registered around seven thousand small providers, classified as such for having less than 5% of the market, around 1.5 million internet clients (Grossmann, 2018b).55

Interestingly, altogether, regional providers are responsible for 22% of the Brazilian fixed broadband market, according to data from Associação Brasileira de Internet 2018 (Brazilian

Internet Association) (Abranet, 2018), with a major role in the increasing number of new connections. For instance, as of May 2018, they were responsible for 74.2% of the new fixed broadband accesses (Grossmann, 2018a). In fact, when considered together, regional providers are usually compared to the fourth biggest internet provider force in Brazil, behind the major telco companies which are NET/Claro (Telmex, Mexico), with 30% of the market clients,

Telefônica (Spain), with 25%, and Oi (Telecom Italia, Italy), with 20%, as of July 2018

(Grossmann, 2018a).

Importantly, beyond these official numbers, the reality shared by specialists who work closely with small ISPs—advising them administratively, juridically and technically—indicates that the number captured by the Brazilian telecommunications regulator are far from that of

55 The Brazilian population is estimated at more than 210 million people in 2019, according to the Brazilian Institute of Geography and Statistics (IBGE). Available at https://www.ibge.gov.br/estatisticas/sociais/populacao/9109- projecao-da-populacao.html?=&t=resultados. 237

reflecting the universe of providers within the country, given that many of them do not pursue a license.56 Lacier Dias, the owner of a consulting company that serves more than 2,700 small ISPs with a range of 200 to 300,000 broadband clients, reports that small providers commonly start contracting a residential internet plan of an ISP, sharing that connection and charging an amount for that, without further formality. As the number of clients increase, these shared networks start to become institutionalized, only then applying for a license and purchasing autonomous system numbers. Lacier Dias gives a sense of this vacuum:

We did a study in the city of Rio de Janeiro, and there is an average of three providers per favela. Rio de Janeiro has 700 favelas, so there are more than 2,100 providers. The overwhelming majority of these providers are illegal. ANATEL is not going to enter the favela to check. And [these providers] do not have autonomous system number. They are doing NAT (…), as if they were companies with 500 employees, and not a company that sells internet access to 500 clients. So they operate on the edge of any organization (...)57

Importantly, these internet providers, devoid of ASNs, cannot interconnect to an IXP, as they are not formally an autonomous system. However, they may become one in the future, so looking at this scenario with commercial lens, Brazil is a very promising internet interconnection market to explore, not only for the huge number of ISPs formally registered today, but also for the potential ones.

56 The need for a license for the provision of internet services is a controversial issue in Brazil. The internet is classified as a value-added service, and according to Article 61 of Brazil’s General Law of Telecommunications from 1997, “Value added service does not constitute a telecommunications service,” (Brasil, 1997, own translation)—a fact that had been already ratified by the Norm 004/1995, issued by the Ministry of Communications two years before (Brasil, 1995). Because of that, why ANATEL should regulate who can and who cannot provide internet becomes a question of interpretation that creates a division between the agency and other sectors of society. Interestingly enough, no license is necessary for an ISP to acquire an ASN and become part of an IXP. 57 NAT stands for Network Address Translation, and is also known as Carrier Grade NAT (CGNAT). ISPs have frequently used such resource to deal with the limited availability of Internet Protocol addresses. With this technique, ISPs share a unique IP with many clients simultaneously, and while the larger internet sees only one IP at use, internally, the ISP has numerous private IP addresses designated (NIC.br, 2014). 238

IX.br is not a business, though, and the fact that it did not charge for any interconnection services until recently helped to attract an impressive number of ISPs to its facilities. On the one hand, at IXPs, small ISPs have the opportunity to collaborate among each other sharing their infrastructure, their routes, and the routes of their clients, giving each other feedback. On the other hand, centralized at one point, they become more easily reachable to content providers. As the Peering Coordinator of a big CDN makes clear about the importance of being present at

IX.br, IXPs means economic viability for CDNs to reach small providers due to its centralizing design:

One cannot peer directly with all [Brazilian ISPs]. They are small, but together they generate a lot of traffic.

Yet, regarding the design of the more than 30 IXPs coordinated by NIC.br in Brazil, it is a political and economic decision to not connect the IXPs to each other. According to Antonio

Moreiras, NIC.br Projects and Development Manager in conversation with the author, this is a way to not compete with telco companies, whose business also relies on selling transport for networks within the country. He also admits that the cost for interconnecting the IXPs would be impeditive for the project. In view of that, each of these nodes facilities needs to generate a substantial level of attractiveness by their own to meet the mindset of CDNs and content providers, and lead them to cache content highly demanded by users. This is where a catch-22 situation is formed.

It is when content is available at IXPs that internet providers become willing to interconnect there. That becomes a challenge in IXPs that NIC.br created with a regional development approach, aiming at deploying local connectivity. If these IXPs do not have yet the traffic considered acceptable for business purposes, content providers refuse to install their

239

servers in these facilities and consequently, networks do not invest in interconnecting to that nodes. This is an important reason to understand the success of IX.br São Paulo and how it attracts networks from other states where there are already IXPs, and theoretically, would be much more cost effective for a network to interconnect, but, for the reasons noted, they ended up being not. On that, Percival Henriques, a CGI.br elected member to represent NGOs, explains how the IX.br São Paulo became a solution to networks located in other regions. He was one of the main advocates for building IXPs in the Northeast; however, he soon noticed that the facilities were perishing without content, and his organization, the Associação Nacional para

Inclusão Digital (ANID) (National Association for Digital Inclusion), ended up investing in another solution to facilitate the arrival of small ISPs to São Paulo:

ANID's mission has been to lower the cost of internet infrastructure and increase the independence of small providers from telcos. And the IXPs, as a neutral space, would be a place where people could meet (…) After PTTs [in the Northeast] were opened, [networks] spent years [exchanging] little traffic between them... So an alternative to reducing the value of broadband [for that ISPs immediately] was to exchange traffic in São Paulo. (...) We rent a place in São Paulo in a data center, convinced NIC.br to put a switch in our rack, and we contracted [transport] circuits from several points in Brazil to São Paulo, something like 20 to 30 circuits. With that, small ISPs [were able] to arrive in São Paulo and interconnect there. We added 80 new participants to IX.br SP because there was already content there, Google, Globo, etc., and in other places there was nobody yet.

IX.br SP was then a second-best option for these regional internet providers. Since then, some other solutions have been implemented to bring content to these non-attractive regional

IXPs for CDNs and content providers. One of the solutions is to share the cost of hosting such contents among the IXP participants, as it becomes cheaper than paying transit for a Tier-1 or

Tier-2 to access that content. This collaborative project has been implemented in Campina

Grande—a very small town whose IXP has become, after such implementation, the fifth in traffic within the country, in front of many capital cities. NIC.br has tried to implement such a 240

solution in other regional IXPs, calling the project OpenCDN, by the time of writing, under an experimental phase in a Northeast capital, Salvador.

Enabling highly demanded content in regions that commonly do not attract big content providers could have as a positive consequence distributing the protagonism of IX.br SP with other regional IXPs. Such outcomes are yet to be seen. As this initiative evolves, it deserves to be closely monitored to help further understanding the private networks strategies that underlie the internet ecosystem.

The dynamics seen in Brazil, which engender and amplify the importance of the IX.br SP for the country’s connectivity, is a sample of how an IXP is a network of relations that require collaboration, and how some players, especially content providers and CDNs and the infrastructure that supports them, emerge with a great power to impact the flow of information on the internet. These companies are at the core of the internet interconnection politics. The concentration outcomes, which result in the creation of a giant internet node in a large country as

Brazil, are just a local dimension of a phenomenon with implications that surpass national borders.

6.2. Globalizing the Local in a One-Way Direction: from the North to the South

When taking seriously the necessity to “learn from the South and with the South”

(Santos, 1995, p. 508), the dynamics visualized within a country as Brazil gain relevance beyond the local, as the most demanded content there is produced by a few global North companies, In fact, according to Julio Sirota, Infrastructure Manager at IX.br:

241

In São Paulo, typically, the three more significative content providers represent 60% of the traffic at the IXP.

As companies, content providers have the discretionary power to decide where to host their CDNs, following the infrastructure available, but also their own economic interests—as shown by the situation in which IXPs were created, but highly demanded content was not attracted. During my interview with Francisco Badaró, the engineer responsible for the interconnection of the unique Brazilian ISP at DE-CIX as of February 2018, he cited Apple as an example of content that was not accessible at IX.br SP, but was highly accessed in Brazil.

Following the actor-network path, at PeeringDB it is possible to see that the Apple’s interconnection strategy in public IXPs is to be present primarily in global North nodes.58 I tried to reach Apple to understand the company’s reasoning to not install CDNs in more global South countries, including Brazil. The answer helps to contextualize the highly privatized values that are embedded in internet interconnection arrangements:

Apple is a very secretive company and I do not have permission to collaborate on many projects.

In this scenario, when content is not locally cached, once a Tier-3 ISP in Brazil receives the request to access Apple content, its routers will define where to send that data packet, analyzing the best route, or the best AS_PATH. IXPs have commonly higher priority, but if content is not reachable in the IXPs where that ISP is connected, the ISP will send the packets through a Tier-2 or Tier-1 transit provider. The fact is that if this content is present in an IXP elsewhere, why not interconnect there? The reasoning is economic: if paying the transport to reach that node, even if in another continent, is compensated by the content that will be available

58 Available at https://www.peeringdb.com/net/3554. 242

through peering agreements without further cost, that becomes a good choice. Associated with resilience and with the goal of having as many routes as possible in their routing tables, autonomous systems operators may interconnect in many IXPs as long as reaching them is economically viable.

This is the moment when leading IXPs, as DE-CIX, become very attractive, not only for the networks that are directly interconnected to them, but the routes that such networks announce and the other networks that they allow to reach. This is essential for the success of an IXP, because it allows networks to be closer one to another, diminishing what is known as laterality, with positive economic feedbacks. As Francisco Badaró explains:

[Comparing DE-CIX to a transit provider] the RTT [round-trip time] is always a bit better, 3, 4, 8 milliseconds... I've had cases of 12 milliseconds of improvement. [Nevertheless] the optimized laterality is the target [to interconnecting there], as this prioritizes traffic from [ISP] to my clients. For example, in the case of a client of mine who has [my company and another company as its ISP, and my competitor is not connected at DE-CIX], in this case [without further configurations, the client’s] traffic will come to me, following the principle of the lowest AS_PATH (own translation).

To further understand the gains in terms of laterality, the main driver pointed by this regional ISP operator from Brazil, Figure 27 is clarifying. Although IX.br has more networks connected to it than DE-CIX, as previously seen, when attention is given to the networks that such networks announce, the latter has more than three times more networks reachable from the networks connected to it than the IX.br (48,366 vs. 14,106). For each of these origin networks announced, numerous prefixes are announced too, and the differences between DE-CIX and

IX.br in this case are even higher (496,909 vs. 105,960), meaning that the connectivity of an autonomous system member of a leading global North IXP as DE-CIX is considerably superior when comparing to a member of a leading global South IXP as IX.br. Indeed, considering the

243

full BGP table as of April 13, 2019, with 772,892 prefixes worldwide, prefixes announced at

DE-CIX correspond to 64.3% of it vs. 13.7% regarding prefixes announced at IX.br.

Unique Prefixes and Origin Networks Announced at IX.br and DE-CIX

105,960 IX.br 14,106

496,909 DE-CIX 48,366

Unique Prefixes Origin Networks

Figure 27. Unique Prefixes and Origin Networks Announced at IX.br SP and DE-CIX Fra. Source: IX.br Looking Glass and PCH Raw Data, March 24, 2019.

Yet on the disparities, from the total of 90,423 ASNs allocated worldwide (Maigron,

2019), as of April 8, 2019, 53.5% versus 15.6% of them are reachable at IX.br and DE-CIX, respectively (see Figure 28) . Considering only the 64,068 ASNs active in the BGP routing system as of April 13, 2019 (Bates et al., 2019), these numbers go up to 22.0% versus 75.5%, respectively.

244

Proportion of ASNs Announced at IX.br SP and DE-CIX Fra

22.0% IX.br 15.6%

75.5% DE-CIX 53.5%

BGP Routing System Allocated

Figure 28. Proportion of ASNs Announced at IX.br SP and DE-CIX Fra. Source: IX.br Looking Glass, PCH Raw Data, March 24, 2019, Bates et al. (2019), Maigron (2019).

Also, considering how many countries are represented among the networks announced at

IX.br SP and DE-CIX Fra, from 238 countries with at least one ASN allocated, 58.4% (N=139) are reachable at IX.br SP versus 94.5% (N=225) at DE-CIX Fra (see Figure 29).

Countries Represented in Announcements at IX.br SP and DE-CIX Fra

IX.br 58.4%

DE-CIX 94.5%

Figure 29. Countries Represented in Announcements at IX.br SP and DE-CIX Fra. Source: IX.br Looking Glass and PCH Raw Data, March 24, 2019, Maigron (2019).

245

Looking at the coverage of ASNs allocated per country at both IXPs, DE-CIX performance is consistent in both the global North and global South. In more than 70% of the countries from both regions with ASNs reachable at DE-CIX facilities, DE-CIX covers 50-100% of all ASNs allocated in that countries. In the case of IX.br, the scenario is completely different.

First, it covers global South and global North countries’ allocated ASNs differently in at levels far from below DE-CIX coverage. For instance, 17.7% of the global South countries vs. 3.3% of global North countries have 50-100% of their ASNs reachable at IX.br. Following that, IX.br ends up being currently able to cover up to 49.9% of all allocated ASNs in countries from both regions.

Table 19. Coverage of Allocated ASNs per Country from the Global North and Global South at IX.br SP and DE-CIX Fra

IX.br SP DE-CIX Fra N % N % Global North 60 100.0% 83 100.0% 50-100% of allocated ASNs per country 2 3.3% 59 71.1% 0.1-49.9% of allocated ASNs per country 58 96.7% 24 28.9% Global South 79 100.0% 142 100.0% 50-100% of allocated ASNs per country 14 17.7% 105 73.9% 0.1-49.9% of allocated ASNs per country 65 82.3% 37 26.1% Total North and South 139 100.0% 225 100.0%

Finally, the advantages of DE-CIX are expressed in its traffic as well. On average, it reaches 4.28 Tbps versus 2.99 Tbps at IX.br, with picks of 6.88 versus 4.21Tbps, respectively.

With this scenario, the jump of Brazilian networks at DE-CIX within a year, from 1 to

17, is better understood. In a competitive communication market, ISPs will seek to neutralize competitors’ advantages and maximize their own as much as possible. As this results in Brazilian

Internet Service Providers exchanging traffic in IXPs beyond national borders, the sovereignty 246

factor embedded at the Brazilian IXP ecosystem, idealized by CGI.br and administered by

NIC.br, is challenged by the sovereignty of the ISPs. This corporate sovereignty is an idea embedded not only in the concept of autonomous systems, but it is generally introjected in network specialists’ vision who see the internet as a communication system without borders— despite the fact that the borders are there, including in the RIRs information distributed by continents, and the statistics about ASNs explored in the present research. As an example of this mindset, below is how Adilson Fiorentino, a BGP specialist, interprets this scenario:

The question of sovereignty, at a more specific level than the sovereignty of states, is the sovereignty of the enterprises themselves. The concept of autonomous system is just that, is whether or not you can send the data from one company to another. Each company, each autonomous system, is sovereign to send or not. From the moment it sends the data, whoever receives it is also sovereign to do as it wishes. Who will control the autonomous systems? This is a philosophical question, as then they will cease to be autonomous (own translation).

The result of such a predominant view is that it limits further discussion on the global internet interconnection market dynamics and the continuous imbalances that they generate.

While such imbalances tend to be understood as a “natural” consequence of both the internet

“free flow” of information and the network effects, by adding countries’ borders to the discussion, the present research provides evidence against such arguments. In fact, the inequalities that arise from BGP routing information are better understood as an evidence of the

“one-way flow” of information that shapes the internet interconnection ecosystem, thanks to resources differently available in the South and in the North, and that tends to be deepened by the market. While it would not be in the public interest to block this flow of information, new channels through which it could flow without being canalized to very few countries from the global North should be deployed.

247

In 2018, the first submarine cable connecting Latin America and Africa started to operate between Fortaleza, Brazil and Kribi, Cameroon. Angola Cables, a telco company from Angola, dedicated to the wholesale market through submarine cable systems, initiated its role as a DE-

CIX reseller in Brazil, offering both transport and the connection at the Frankfurt facilities altogether (Angola Cables & DE-CIX, 2018). With that, DE-CIX Fra, the largest IXP in the world as they advertise it, becomes even closer to ISPs miles away.

Clearly, this new path between Africa and Latin America—celebrated for the expectations that it will leverage new flows that do not need to pass through the U.S. and other

Western countries—will contribute to maintain some path dependencies instead of creating new possibilities if other sociotechnical aspects are not developed. These include more competitive markets and better local infrastructure such as optical fiber links connecting the cities within

African and Latin American continents, data centers, etc.

Resellers are substantial actors in the DE-CIX design. They increase the IXP customers basis and, in the case above, facilitate the interconnection with virtual highways tailored to reach that IXP from different points in the world—with similarities to what ANID has done in Brazil to populate the IX.br SP with ISPs from the Northeast region, and what other ISPs currently do as well. As Francisco Badaró explains based on his experience with ISPs in Salvador, Tier-1 companies explore this market following different business models:

In the business model with Telefónica International [TIWS] (...), TIWS absorbs the cost of the IX’s port. TIWS has reserved a 100-Gbps port on DE-CIX and made a circuit where it transports its customers from Brazil to there. And we can get [the transport] out of Fortaleza, Sao Paulo, Rio and Salvador. To [connect to] DE-CIX, we pay the transport [and the port] to TIWS. [Following another model,] Level 3 charges the transport [to its customers] and lets us negotiate with the IX, which is the default.

248

Additionally, regarding other forms of design centralization, that ISPs read as gains in connectivity, unlike IX.br design in which the more of 30 IXPs are not interconnected, some of

DE-CIX international locations interconnect, offering its customers the possibility of interconnecting at Frankfurt even if they are a member of DE-CIX Moscow, for example—a product that the company calls “GlobePEER Remote” (DE-CIX, n.d.-a). And considering that, according to one of the company’s annual report “The goal for the coming years is to enable new and existing ecosystems in all major telecommunication markets across the globe”

(DE-CIX, 2017, p. 6), this kind of product tends to grow.

On the one side, these options of international resellers and remote interconnections could be seen as a dimension of promoting better inclusion and connectivity for the global South, once DE-CIX and other IXPs in the global North become more accessible to networks that want to interconnect internationally and become less dependent on Tier-1 and Tier-2 links. On the other side, what this phenomenon portrays is a scenario that tends to more data traffic concentration and more centralizing effects on internet infrastructure, moving now from global

North Tier-1 networks to now global North giant IXPs; from any idea of utopian distributed internet to an internet with notable points of centralization.

As Nanette Levinson and Derick Cogburn argue about privatization in internet governance “While this is an economic issue—an excessive burden placed on private information intermediaries to carry out this governance, and a need for economic freedom for the companies providing basic Internet infrastructure—it is also an issue of legitimacy and public accountability” (Levinson & Cogburn, 2016, p. 220).

249

Using global communication lens to look at this picture, it seems important to ask: is that the kind of internet that we want?

7. Final Considerations

By addressing the consistent concealment of the South in the internet infrastructure analysis, this chapter reveals that there are substantive differences when comparing routers’ communication in the biggest IXPs in the global North and in the global South. It exposes drivers that underlie internet infrastructure and produce a one-way flow of information between

North and South placed on uneven availability of infrastructure in both territories and a continuous interdependence among their interconnection facilities. It is because IXPs in the global South do not count on highly demanded content rooted in the North, and consequently do not attract as many networks with worldwide reach, that global South Internet Service Providers opt to interconnect at global North IXPs. With that, these providers become more competitive, because of the BGP design, and end up contributing to consolidate these IXPs’ rich mesh of international networks, and creating a virtuous or vicious circle, depending on the perspective of who is looking at the phenomenon, from states or autonomous systems sovereignty; from an IXP that has been successful in becoming “global” through the globalization of its localism, or an

IXP that has been not.

Leading IXPs are deeply dependent on the poor infrastructure resources in other regions, and have symbiotic relations with content providers. As the latter have become highly concentrated, the same can be said about public internet interconnection facilities as IXPs, evidenced by giant internet nodes in the Norths, including the ones situated within the South. 250

If the logic behind the giant IXPs is not commercial, solutions based on collaboration may emerge. The OpenCDN project is an example of that. If, however, there is a “logic of accumulation” embedded in such artifacts, the escalation of the concentration effects is more likely to be expected.

Shoshana Zuboff wisely points that “The logic of accumulation organizes perception and shapes the expression of technological affordances at their roots. (…) [It] produces its own social relations and with that its conceptions and uses of authority and power” (Zuboff, 2015, p. 77).

Considering the historical relations between the North and the South, this is one more factor that shapes the coloniality of power, where economic dynamics beneficial to global North companies are sustained and intensified by the global South milieu. On the other hand, since interdependencies are entangled with late and peculiar processes of development and coloniality values, collaborative projects aiming at distributing global North content only do not break with this logic, and become one more example of the South thinking with a North lens. It is hard to imagine systemic changes if there is no room and demand for local content in the South.

Learning with the past, it is worth noticing how despite differences in regard to technologies and the new actor-networks that have been formed around the internet, long- standing patterns of inequality arise from the contemporary communication medium materiality.

As a reminder of previous relations of power, this excerpt from the MacBride report is elucidative:

The critics from the developing countries have found, by experience [,] that the theory of "free flow" is invalidated by the overwhelming preponderance of information circulated from a small number of industrialized countries into the huge areas of the developing world. In order to be really free, information flows have to be two-way, not simply in one direction. The concentration of news agencies, telecommunication facilities, mass media, data resources, manufacturers of communication equipment in a small number of highly

251

developed countries does, in fact, preclude any chance of a free flow between equals, a democratic exchange among free partners. (UNESCO, 1980, p. III-6)

The concept of “free flow” in global communication is as utopian as the image of a completely distributed internet architecture and should be addressed more seriously in policy and internet governance fora. It is true that the current one-way flow of information is generated by demands of consumers from the South and the features of giant content platforms from the

North, but there is nothing “natural” about that, and alternatives should be addressed in a collective way, considering the societies that we want to live in. Differently from the past, the one-way flow of information has more intricated consequences in that personal data is being appropriated to generate new realities based on data associations that feed companies’ profit and state control (Haggerty & Ericson, 2003; Rouvroy & Berns, 2018; Zuboff, 2015).

Alternatives such as China’s WeChat do not solve the problem from the standpoint of a society that has privacy as a collective value, given that such a platform, while local, adds another kind of centralization, in this case within a state whose strategy to ban international platforms maximizes its potential concentration. In this context, content solutions that are locally based, and have interoperability with other platforms but do not allow data governance concentration, could perhaps be a response to the dilemmas seen above on the level of internet infrastructure. Whether this is a real possibility for the future, considering the disseminated desire for control through personal data, is difficult to answer.

In the next chapter, I will analyze further consequences of the internet interconnection politics and the globalization of localisms unveiled here, showing that they go beyond affecting the internet architecture functioning and the general internet flow of information. Despite the neutrality values on which IXPs build themselves, promoting competition values among

252

autonomous systems, for their centralizing design, giant internet nodes end up being opportunistically used for supporting obscure dynamics in sociotechnical networks of relations, in which flows of information are redirected for surveillance practices.

253

CHAPTER 6

IN SEARCH OF BODIES: SURVEILLANCE AND CONTROL AT INTERNET EXCHANGE POINTS

If bodies and embodied spatiality mediate cognition and social ordering in complex and interdependent ways, then we cannot simply leave bodies and spaces behind as we enter the networked information age.

—Julie F. Cohen, Configuring the Networked Self (p. 41)

This chapter seeks to examine a state surveillance case conducted by the

Bundesnachrichtendienst (BND), the German Foreign Intelligence Agency, at DE-CIX

Frankfurt. The analysis shows how the state converts the IXP, centralizing points on internet architecture, at a “centre of appropriation” of data as part of a surveillance assemblage that serves the state interest for control. It shows how the German state relies on controversial interpretations of the laws to extend its associations as a macro-actor and counts on the collaboration of DE-CIX, a private company. The chapter makes the case that to advance our understanding on government surveillance, data packets circulating over the internet need to be understood as “virtual bodies,” with intrinsic characteristics connected to real-selves from a non- virtual world. As a result of that, the so-called massive surveillance is understood not only as highly distributed as the word “massive” implies, but also focused on specific selves.

254

1. Introduction

This chapter aims to shed light on mechanisms of state surveillance through internet infrastructure. Although it is common to find in the literature the term “government surveillance”

(DeNardis, 2014; Marthews & Tucker, 2017; Semitsu, 2011), the term state surveillance will be privileged here, as it adds more emphasis to the robust and perennial aspects on which surveillance practices rely.

In ANT terms, the state is a macro-actor that differs from micro-actors, not in nature or essence, but in the volume of associations that it is able to create and sustain with other actors, which consequentially defines its relative dimensions, as a macro- or micro-actor. In Michel

Callon and Bruno Latour’s words, it is the enlistment of “bodies, materials, discourses, techniques, feelings, laws, organizations” and “the variations in relative solidity and durability of different sorts of materials” (Callon & Latour, 1981, p. 284, emphases in the original), that shapes a macro-actor. On the basis of that is the assumption that actors are networks, which need to be understood in their relations. The state can be understood then as a network of actors kept in equilibrium due to mechanisms that it controls to maintain those relations and associations stable. Regarding such mechanisms, the authors explain:

Two actors can only be made indissociable if they are one. For this their wills must become equivalent. He or she who holds the equivalences holds the secret of power. Through the interplay of equivalences, hitherto scattered elements can be incorporated into a whole, and thus help to stabilize other elements (Callon & Latour, 1981, p. 293, emphasis added).

Holding the “equivalences” is then key to preserve the cohesion of a network of actors.

This may signify an exchange of money and services, as well as an exchange of interests in being more profitable, on the one side, and more business-oriented (also generally and wrongly 255

associated with being more “efficient”) on the other. In the context of neoliberal states, which in

Foucault terms are marked by a propensity towards the rule of political power by the tenets of the market economy (Foucault, 2008, p. 181), scholarship reflecting on state and security shows how tenuous the line becomes that separates the macro-actor from a relative micro-actor, in this case a state and a private company. According to Bruno Cardoso’s research on the Brazilian security policies to host mega-events such as the Olympic Games and the World Cup, “The construction and operationalization of large security technology systems embed these businesses

[security technologies suppliers] into the government's own governance and functioning architecture, with a role that can go beyond merely providing services or materials” (Cardoso,

2018, p. 94, own translation). As an expression of the neoliberal state, he continues, “The State is not only hybridized with companies, but also builds on the model of the company” (Cardoso,

2018, p. 103, own translation).

Based on these facts and premises, three focuses will be given in this chapter. First, at this point there should be no reason to think of states and companies as opposed actors; in many instances, they are actors in relation and their tactics and modus operandi may overlap. Thus, the concept of surveillance capitalism that has been built looking at companies such as Google and their strategies to collect, mine, analyze and monetize personal data and modulate behavior

(Zuboff, 2015, 2019) should not be disregarded. Neoliberal states tend to appropriate private strategies and business managerialism, and datamining techniques are out there to be explored for different purposes. The question to be answered is what kind of association has been formed among private and public actors.

256

In a normative statement, Callon and Latour suggest that in the study of associations, “the transformation of weak interactions into strong ones and vice versa” (Callon & Latour, 1981, p.

300) is an important moment to look at given that this is the moment in which changes in the actors’ dimensions occur. If in the case above on security policies in Brazil, the association can be said to be based on a customer-provider relation, what is it that is formed between the German state and the DE-CIX Frankfurt? DE-CIX’s growth as a private company exploring the interconnection and peering global markets has made it one of the leading nodes of the internet, attractive not only for internet networks interested in improving their businesses. Its design and location have called the attention of the German intelligence agency to set up surveillance technologies at its facilities as part of a monitoring strategy focused on the large amount of national and international data circulating at that IXP. What has been the basis for the state to exploit such internet infrastructure artifact?

Second, discussing state surveillance, Kevin Haggerty and Richard Ericson’s concept of surveillance assemblage contributes significantly to understanding the mechanics of surveillance.

Influenced by Deleuze and Guattari (1987) and Mark Poster (1990), they argue that the state, guided by the desire of control, will create breaks into flows, and for that they need “both spaces of comparison where flows can be rendered alike and centres of appropriation where these flows can be captured” (Haggerty & Ericson, 2003, p. 607). For the authors, with such surveillance mechanisms, a “data double” individual is created, a “pragmatic” and discrete one, “toward which governmental and marketing practices are directed” (Haggerty & Ericson, 2003, p. 613).

What Kevin Haggerty and Richard Ericson’s analysis fails to incorporate is the need for embodiment of this individual. The pragmatic self isn’t disembodied, and sometimes it is

257

reimbodied based on characteristics of thousands, even millions, of other people that apparently have similar characteristics. This leads to reflect on the fact that surveillance tends to affect different individuals differently, but the implication is even more noteworthy. Although there may be less information regarding this, the motivations for surveilling and the techniques of surveillance may also be already embedded in the assumption that some bodies need to be surveilled more than others. Thus, understanding on which bodies the German state focuses its search is also relevant.

Finally, the macro-actor has greater capacity of black-boxing things as richly explained in the excerpt below:

The more elements one can place in black boxes - modes of thoughts, habits, forces and objects - the broader the construction one can raise. Of course, black boxes never remain fully closed or properly fastened (…) but macro-actors can do as if they were closed and dark [,] (…) macro-actors, to say the least, do not have to negotiate with equal intensity everything. They can go on and count on a force while negotiating for another. If they were not successful at that, they could not simplify the social world. In mechanical terms, they could not make a machine, that is hide the continued exercise of a will to give the impression of forces that move by themselves. ln logical terms, they could not make chains of arguments, that is stabilize discussion of certain premises to allow deductions or establish order between different elements (Callon & Latour, 1981, p. 285, emphases in the original).

Inside a black box are “those things whose contents have become a matter of indifference” (Callon & Latour, 1981, p. 285), thus in talking about black boxes “we are dealing with the creation of asymmetries” (Callon & Latour, 1981, p. 285), including asymmetries of knowledge about procedures and practices. That is a crucial aspect to include in discussions about state surveillance through the internet. Once internet surveillance mechanisms are black- boxed, they are naturalized and generally unquestioned. Yet the historical moment that we are living in points to a not-yet sealed black box.

258

Laura DeNardis has problematized the turn to infrastructure as a form of control in the internet (DeNardis, 2012; DeNardis & Musiani, 2016). In the author’s words, “Internet governance technologies not only embed political values in their design and operations but are increasingly being co-opted for political purposes irrelevant to their primary Internet governance function” (DeNardis, 2012, 721). This kind of assessment in the scholarship points to a scenario of a “leaky” black box (Callon & Latour, 1981, p. 285), which in the present research shows transformations in course and open terms. As already warned, “Shared interconnection points can serve as centralized points of control for government surveillance and censorship or concentrated targets for Internet disruptions” (DeNardis, 2014, p. 224). Understanding this black-boxed phenomenon is also part of this chapter’s main inquiry.

2. Methods

The analysis that follows is based on local media coverage, and opinions of DE-CIX’s spokesman for the case, Klaus Landefeld—in the news, a conference, a TED Talk, and a parliament testimony—content primarily available in German and translated into English for the analysis.

The Landefeld’s opinions and testimony will be taken as legitimate sources to understand the modus operandi of the implementation of a surveillance state. In relying on his narrative for such intent it is important to problematize Landefeld’s primary role representing a company concerned with its reputation as a giant IXP and interconnection facilitator worldwide. In my personal conversation with Landefeld at the DE-CIX Summit Frankfurt, in Dreieichenhain,

Germany—an event promoted for DE-CIX customers— reputational concerns with the case of 259

German state surveillance were confirmed when inquired about. Such concerns can certainly generate intentional or unintentional self-censorship.

Nevertheless, for the purposes of understanding the kinds of associations in an actor- network, while the tensions between DE-CIX and the German State materialized into a lawsuit filed in 2016 may be interpreted as a marketing strategy to avoid shaking DE-CIX leading market position, these tensions exist due to differences in arguments that deserve attention.

Moreover, the fact that Landefeld’s discourse is coherent with the general engineering narrative on good practices that sustain (the need of) an agnostic internet is an important analytical element.

Yet supporting DE-CIX position, a legal expert opinion given by the Former President of the Federal Constitutional Court, Hans-Jürgen Papier, requested by the company and published as a legal article was analyzed. Additionally, the testimony to the Parliament of Hans de With, former Chairman of the G-10 Commission (whose functions will be further explained ahead), is also important sources of information. They were all translated into English for analysis.

Finally, I conducted personal interviews with Klaus Landefeld and one of the DE-CIX attorneys following the case, Simon Assion.

In accessing this material, I situate myself as a researcher whose main goal is to leverage the public understanding of internet interconnection politics, shedding light on public interest issues regarding surveillance of public actors via private infrastructure. As already stated,

“While the government, via surveillance, know more and more about what its citizens are doing, its citizens know less and less about what their government is doing, shielded as it is by a wall of

260

secrecy” (Greenwald, 2014, p. 208-209). Problematizing surveillance mechanisms is also a way to show the need for more transparency in the states’ surveillance actions and control.

It is not my goal to assume any other position than my own, a Brazilian citizen interested in unveiling flows of information constrained by laws, judicial orders, political agreements, and technological artifacts that, despite being circumscribed to Germany, have impacts in internet users from other countries as well, as a result of autonomous systems’ sovereignty in determining the flow of information online.

3. De-Black-Boxing Surveillance through Internet Infrastructure

In March 2015, Klaus Landefeld was before the Geheimdienst-Untersuchungsausschuss, the German Intelligence Inspection Committee, for a hearing about the German Foreign

Intelligence Agency Bundesnachrichtendienst (also known as BND) tapping of internet traffic at the DE-CIX facilities. At the same session, Hans de With, former Chairman of the G-10

Commission, provided the second testimony. This commission is composed of four members appointed by the Parliament and its main responsibility is to define whether or not and under which conditions Article 10 of the German Constitution (also known as German Basic Law), which guarantees the privacy of “correspondence, posts and telecommunications,” can be limited

(Germany, 2014).

In 2013, as a consequence of Snowden revelations, the G-10 Commission head has confirmed surveillance operations at the DE-CIX, stating, according to news reports, that around five percent of data traffic is used by German agencies for strategic intelligence purposes

(Ermert, 2013). In 2015, a leak exposed the German authorities to new public scrutiny, with 261

more details on how the BND has tapped into the Frankfurt IXP since 2009 in collaboration with the U.S. National Security Agency (NSA) (Meister, 2015). And although the IXP case is the focus of the present chapter, before analyzing it I propose to look at important contextual information about another German State surveillance event that precedes it.

3.1. German State Surveillance and Internet Service Providers

Klaus Landefeld’s experience as a businessmen in the very beginning of the internet brings important background on disseminated surveillance practices towards information circulating online. After founding one of the first Internet Service Providers in Germany in 1993, he started to operate a Tier-1 backbone network in Europe, which served seventeen countries in the 2000s. He explains that specially after September 11, regardless of the country, the company would have discussions with law enforcement agencies and secret service who were interested in understanding “How to get internet data? How can this work?” (Landefeld, 2015, p. 7, own translation) for collaborations. He mentions the first search that he experienced in 1993 and how data collection was made: “At that time, everything that had been found had to be printed out.

That has produced whole trucks of paper at that time at the state criminal police in Hessen”

(Landefeld, 2015, p. 6, own translation). The way that he would deal with this kind of approach was in terms of duties. In his words, “But of course, if you are a provider then you have to support law enforcement authorities, that is obvious. That is a legal obligation, but of course it is also in the interests of society” (TEDx Talks, 2015, own translation, emphases added).

At DE-CIX, Klaus Landefeld explains that informal conversations took place with government agents in 2008, when BND representatives approached the company to install 262

surveillance apparatuses to wiretap a number of different networks. Interestingly enough, that year points to a likely encroachment stemming from another German state surveillance operation known as Operation Eikonal, which brings important contextual information to understand the posterior onrush at DE-CIX.

According to news reports released in 2014, this operation was conducted by the German and the U.S. intelligence agencies, BND and NSA, to wiretap communication traffic passing through the main German telecommunication company, Deutsche Telekom. Broadly, the operation was part of NSA RAMPART-A program, which, according to NSA confidential documents accessed by journalists, involves partnership with countries “that assist NSA in tapping fiber-optic cables carrying international and national telephone and internet traffic on the territory of the countries concerned” (Geist et al., 2014).

Operation Eikonal started in 2004, intercepting telephone traffic, and then internet traffic in 2005 (Mascolo et al., 2014). For that, according to a witness with a technical background, splitters were installed at the Deutsche Telekom’s cables at one of the data centers in which the telco is present in Frankfurt, and data was copied and sent for further analysis to the surveillant agents, following the rules of a secret agreement between the two agencies (Biermann, 2014). On the basis of this collaboration was NSA’s expertise and technologies, on the one hand, and

Frankfurt’s data hub that attracts great amount of traffic and ended up being eventually accessible to BND, on the other. The interpretation of the law was key for BND’s access. A G-10 commission was issued allowing BND’s surveillance practices, and confirmed by the Federal

Chancellery, BND’s superior body, after the telco’s consultation (Biermann, 2014). Also, the

263

telecommunications provider received 6,000 Euros monthly for giving access to its servers, the report says (Mascolo et al., 2014).

For BND, NSA partnership was a way to learn its techniques (Geist et al., 2014;

Mascolo et al., 2014) as well as to keep pace with security needs, through NSA’s “partners” and

“know-how,” as informed in a classified memo of the agency (Baumgärtner et al., 2015). The

NSA seems to be viewed generally as a necessary evil, and a welcome partner in a scenario that involves national security concerns.

Journalists describe BND and NSA agreements as a business relationship, which includes

Germany’s participation in huge U.S. surveillance operations such as the PRISM, as well as eventually no payment for NSA services, and also physical facilities benefits (Mascolo et al.,

2014). On the latter, the most well-known is the Bad Aibling Station, located in the state of

Bavaria. Built in the midst of the Cold War, it “was established in 1952 as a Field Station belonging to the Army Security Agency (ASA), the organization that provided special communications support to the U.S. Army” (Schindler, n.d. p. 5). During its operations, it was owned by the U.S. Department of Defense and the U.S. Army (Schindler, n.d.), and has been referred as a “listening station” (Dagmar Dehmer & Haselberger, 2015; Mascolo et al., 2014), which was turned over to Germans in 2004—despite the fact that Americans and Germans are reported to have continued work there together in secret (Baumgärtner et al., 2015; Mascolo et al., 2014).

The controversial Eikonal Operation activities then involved data collection at the data center in Frankfurt—where Deutsche Telekom is co-located and intelligent agents rented two rooms according to a witness (Biermann, 2014)—and data dispatch to the Bad Aibling Station

264

for further analysis—a moment when data should be converted into information. With such procedures, this operation controversially provided the U.S. with access to massive data circulating in German territory, including German citizens’ information.

It is reported that, in 2003, both agencies established a “Memorandum of Agreement” aimed at conducting global surveillance activities under the argument of September 11, but in which both Americans’ and Germans’ data should not be monitored (Baumgärtner et al., 2015).

The BND implemented filters to exclude Germans’ phone numbers and e-mails from the massive data; however, the mechanisms were not completely effective (Geist et al., 2014; Mascolo et al.,

2014). Indeed, this kind of separation is not technically feasible, and although Americans were supposed to not use such content, NSA confidential documents suggest that expected good practices on no data collection of host countries are not completely guaranteed and may have exceptions (Geist et al., 2014).

In fact, the U.S. interests were first thought to be primarily Russia information (Mascolo et al., 2014) and terrorism threats, but in 2005 the BND found out that content on French authorities, and local aerospace companies such as the European Aerospace and Defence Group

(EADS, currently Airbus) and the Eurocopter Group (currently Airbus Helicopters), were also being mined using specific search terms (Dagmar Dehmer & Haselberger, 2015; Mascolo et al.,

2014). On the other hand, journalists report that BND information points to the fact that, accidentally, the U.S. Secretary of State Hillary Clinton was intercepted by German authorities involuntarily “as a bycatch,” (own translation) and other authorities from initially friend countries, such as Turkey, were also bugged (Dagmar Dehmer & Haselberger, 2015). This shows

265

the weak control that one can have after having access to populations’ massive amount of data, and the limit of trust in macro-actors’ associations based on surveillance interests.

In 2008, a BND internal memo relayed that there were technical difficulties to separate communication traffic that was and was not protected by law (Mascolo et al., 2014). However, despite initial report that the NSA-BND partnership had ended that year, Edward Snowden’s revelations in 2013 pointed to continuous collaborations despite legal restrictions and sensitive sovereignty relations with the U.S., and concerning spying activities for German public opinion conducted by NSA on Western European countries (Baumgärtner et al., 2015). Indeed, as a result of the parliamentary NSA investigative committee in 2014, results point to more than forty thousand suspect NSA search terms, despite previous BND testimony arguing that the partnership between the intelligence agencies and its practices were legal (Baumgärtner et al.,

2015), and thus not a risk to German sovereignty.

3.2. DE-CIX Surveillance

3.2.1. Flirt and Possibilities

The German state approach at DE-CIX in 2008 seems to be a new step in internet surveillance, initially directed to networks themselves. IXPs’ design and the possibility of reaching many networks at once would be attractive enough for that purposes. The fact that DE-

CIX was not only a central access point to Internet Service Providers, but also the biggest

German access point, certainly increased that attractiveness. Also, the fact that Frankfurt is part of a physical ring which numerous international lines cross, and which providers build lines

266

between—for instance, Amsterdam and Frankfurt, Frankfurt and Moscow—and exchange traffic at DE-CIX (Deutscher Bundestag, 2015, p. 96), the IXP concentration values increase because it changes the flow of information that could be otherwise. As Landefeld explains:

“(…) [R]ings from different regions come together in Frankfurt. And here, in fact, the abstract network logic of the Internet follows also the physical structure. That means, in Frankfurt a lot of traffic, that means a lot of transit traffic is exchanged. This means that only part of the traffic is German traffic. There are of course a lot of international traffic, and you can get it very well at this point. You would not see it in Stuttgart, or you would not see it in Dusseldorf or anything like that. You only see it in Frankfurt, because only here the nets have interconnected in this form. For example, traffic from Eastern bloc countries to the West is halted, and [traffic] from the Arab world, but also from other regions, which is mainly being transported in Frankfurt, is of great importance” (Deutscher Bundestag, 2015, p. 83).59

At another point of the document, after hearing an explanation about the amount of data that circulates through the DE-CIX infrastructure, the session chair, Dr. Patrick Sensburg, admits: “Of course, with this huge volume of data, it seems to me to be very interesting for many to be able to look at it, especially if it is the biggest node point” (Deutscher Bundestag, 2015, p.

13). Thus, despite the emphasis on the neutral role of IXPs in autonomous systems arrangements widespread in IXPs operator discourses and network engineering scholarship, one should not neglect that as nodes in the circulation information infrastructure of the internet, IXPs may become convenient instruments for other purposes than the ones that they are designed for.

When asked about DE-CIX capabilities to see the data that is circulating to its customers,

Landefeld’s answer is clear: “Of course. - I mean, as an operator, of course, you could theoretically access everything. But of course that’s in the nature of things. Every line operator could in theory always access all traffic on this line” (Deutscher Bundestag, 2015, p. 15, own translation). The IXP is responsible for managing its local network as a whole, connecting and

59 This and all citations to this document are of own translation. 267

disconnecting, physically and logically, networks to its switches. On the other hand, customers’ access is allowed to their own data only, which is directed to their ports in the edge switch to then reach their routers.

By design, there are different possibilities for opportunistic actors to intercept communication at IXPs. And while it is always possible to compromise devices with viruses and other malicious programs—and that should not be disregarded—Landefeld warns (Deutscher

Bundestag, 2015), what German state practices analyzed here show is the option for formal negotiation with a network operator, as seen in the Operation Eikonal, and with an IXP operator, as in the DE-CIX case, where many networks can be reached at once.

Interestingly, the informal conversations in 2008, or “preparatory talks,” as interpreted in the hearing (Deutscher Bundestag, 2015, p. 49), occurred before any official order was issued.

They wanted to know how an IXP works. In Landefeld’s words:

Also there was, for example, the function of the node, the structure of the node, how it actually works, etc., Similar to the content of [this] meeting [the hearing]. So, clearly these talks were used to get information, what actually needed or which lines, which providers are there and where they are, etc. (Deutscher Bundestag, 2015, p. 49).

This exploratory approach from government officials indicates their dependence on internet businesses to acquire technical understanding about internet infrastructure. On the other hand, according to the DE-CIX representative, it was a one-way conversation in which the agents envisioned intercepting the IXP as a whole. In Landefeld’s words, they “tried to go to the node at all. And our understanding was that you had to specifically call individual lines and could not somehow go to the node and say: ‘Now I would like this line today, that line tomorrow’, etc., etc.” (Deutscher Bundestag, 2015, p. 59). This led the company to look for support, arranging a meeting with the G-10 Commission, but at that point when all they had were conversations

268

previous to any implementation, the feedback was that they could not help. At the same time, after the company raised concerns in implementing the surveillance measures, they were invited to have a meeting in the German Chancellery, to which BND responded, being required to continue to collaborate. In Landefeld’s words:

Basically, even before the order has been issued, we have just tried to take action against this technical discussion or, for example, to talk to the G-10 Commission and say: (…) We have those and concerns. - And there, we were just then summoned to the Chancellery, and we were told very clearly that we [had] to talk before anyone has an [order, an] arrangement with anyone, not even with the G-10 Commission or the Parliamentary Control Panel, [and that] we must not talk to anyone about any measures that [were been] planned (Deutscher Bundestag, 2015, p. 47).

3.2.2. Engagement and the Role of the Law

Following the first contacts, in 2009, DE-CIX received a G-10 order focusing on specific lines and networks, including German and international ones. The implementation of such an order and its consequences is highly opaque, as there are many ways to do it, and concerns of being responsible for opening this black box to the public. Landefeld makes it explicit: “Of course there are different technical procedures as well as different implementations. (…) I do not know if this can be discussed so neatly in the public part [of the hearing] now. I have my doubts”

(Deutscher Bundestag, 2015, p. 74). Nevertheless, he generally synthesizes two ways to tap a network, describing distinct design options—mirroring a switch port or tapping an optical fiber cable—and the levels of cooperation necessary between an operator and an opportunistic actor.

For its public interest relevance in making a layer of the politics of interconnection visible, I transcribe it below:

So basically (…) there are actually these two possibilities. The one would be to work in cooperation with the provider. Then, in principle, put a normal line to the provider, as 269

described earlier. You want to turn it on at a switch, and the provider would then give you a copy of the traffic completely on it. It is then the task of the BND to filter that, first time to add the regional filter and something [distinguishing German and non-German traffic to comply with national law]. But that already works at the service, so to speak. So, in the rarest case, this technique is actually with the provider. It’s not worth it either. It would of course be better to first have only [access to] the line and then somewhere to have a central operating room, where to do that [filters] then. That makes more sense. That’s how it is done. (…) The second option would be to insert exactly [a] coupler into a glass fiber. Then you need a reinforcing element [because it is a sensitive action]. That must be very, very close to it. That is, there is a little active technology then connected with [the fiber], and this is then also given directly to [another] line. This is actually a bit more complicated because you never know exactly what’s on that fiber. (…) [A]nd if you actually want to completely derive [the content] what you would do with a bending coupler, you must actually again a broadband (…), then need a dark fiber back there and so (to redirect the data to your own equipment] (Deutscher Bundestag, 2015, p. 74, emphases added).

First, it is worth mentioning the role of a “bending copper” cited above, as they are kinds of non-humans interceptors that create new paths for data to go through. Landefeld explains that it differs from a “splitter” that will cut the fiber and can be identified if fiber tests will be conducted at any point. It is more sophisticated as it bends the fiber instead as he explains:

“[T]he glass fiber is bent over the critical angle of total reflection, and then part of the light exits, so to speak on the outer wall again, and you can catch that as well. But then you have to reinforce that immediately. Otherwise you cannot do anything with it. This is, so to speak, the outfeed technique, if the fiber is not allowed to be cut” (Deutscher Bundestag, 2015, p. 75).

This description captures the materiality of the routes built with optical fiber tubes that permit data packets to traverse from one point to another on the internet, and that any redirection is not only logically but physically promoted. In this way, considering the fact that these packets are pieces of complete communication sessions broken in a packet-switching process, Landefeld suggests that it is unusual that correlated packets—parts of an e-mail for example—will follow different paths. Nevertheless, even if they traverse different lines, it is always possible to reassemble them with the information contained in the packets header. And other mechanisms

270

can be followed to reassemble parts of communication that come from different IP addresses as occurs in the case of a website reassemble, in which the parts may come from different parts of the world (Deutscher Bundestag, 2015, p. 77).

Second, from the hearing, it is not possible to know for sure which strategy was used by

DE-CIX and BND to tap the lines crossing the IXP. However, Landefeld provided much more details throughout the session on the process of mirroring a switch port. Regardless of that being the current procedure of wiretapping at DE-CIX, this content contributes to a better understanding of the possibilities of surveillance at any IXP, and deserves special attention.

With the technical support of the IXP operator, the opportunistic actor may mirror a port through which data of a certain network traverses and can be copied. Parallel to that, the interceptor will have to have access to its own port with a better capacity and the same or better speed transmission line to receive the data. As Landefeld explains:

If you want to mirror a port, so the data must indeed go somewhere. That is, this complete duplicate created there needs at least the same turn-on speed on any other line again to dissipate it (…) [Also,] your diversion port must always be larger than the port you want to monitor, because both directions have to be transferred into the same direction, so to speak if you want to monitor a 10 Gigabit port, you need to have 20 [G]igabit attachment capacity (Deutscher Bundestag, 2015, p. 16).

Importantly, Landefeld also makes clear that these data mirror procedures are facilitated by the affordances of telecommunication devices in the market built to comply with the U.S.

Communications Assistance for Law Enforcement Act (CALEA). This is a law passed by the

U.S. Congress in 1994 with a focus on creating the conditions to monitor digital telephone calls, and modified in 2005 by the U.S. Federal Communications Commission (FCC) to include

Internet Service Providers and Voice over IP services (VoIP), as the organization Electronic

Frontier Foundation explains (EFF, n.d.). CALEA requires that telecommunications equipment

271

“have the necessary surveillance capabilities to comply with legal requests for information”

(FCC, 2019). In Landefeld’s words:

Basically, it is like that - you have to say that now; maybe negative, but you have to say very clearly -: Basically, each switch supports - - Every major switch supports such a function, also has so-called CALEA functionalities, so everything is legal interception. So this functionality, that something can be switched, is basically given everywhere. (Deutscher Bundestag, 2015, pp. 16-17).

Put another way, internet interconnection devices already embed surveillance affordances that impact the values of global communication with the extraterritorial effects of laws passed in powerful states as the United States, also known as a “surveillance state” (Edmundson et al.,

2016). For instance, in an interesting RFC, Cisco, a leading router vendor, details its

“Architecture for Lawful Intercept in IP Networks” (Baker et al., 2004), relating the open formalization of this chain of effects, and one more “globalization of a localism” embedded into the internet interconnection infrastructure.60

Interestingly, the main concern of the chairman in Landefeld’s hearing was to investigate whether intelligence agencies from other states (specially the United Kingdom and the United

States) could have access to DE-CIX and the Germans’ information that circulates there. By testing hypothetical possibilities of tapping without the DE-CIX cooperation, the inquirer would lead Landefeld to constantly reaffirm the level of security and procedural monitoring that DE-

CIX has.

Some of the arguments that the witness used to legitimize DE-CIX security are: multiple

(“six or seven”) fiber providers in Frankfurt and “no single points of failure;” continuous

60 On that, a network operator shared his resignation in an interview: “Technically speaking it is impossible to avoid veiled or explicit interception. The great proof of this is that in Cisco this [CALEA] software component was hidden and was revealed by Wikileaks. After the Wikileaks’ revelation is that Cisco has positioned itself about CALEA. Until then, I, a Cisco system administrator, did not know.” 272

measurement of attenuation in their optical fiber network to be alerted of any changes; multiple

(eighteen) and “highly secure” data centers, with “multi-level access controls;” “own equipment;” “locked racks [with] no physical access for third-parties;” “four-eyes principle,” with four people focused on security; remote access to the network via VPN (Virtual Private

Network) only; and the work “closely with the Bundesamt für Sicherheit in der

Informationstechnik (BSI) (German Federal Office for Information Security)” to improve procedures and make them failure safe (Deutscher Bundestag, 2015).

Importantly, in consistently reaffirming the DE-CIX security, Landefeld exposes its primary flaw. First, considering the procedures of mirroring switch ports at an IXP—which involves the need for the IXP to make ports available, connect the cables and logically manages the switch functions for deviating the flow of information through the new port—counting on the technical support of the IXP is the easiest way to afford that goal. As exposed in the hearing:

Chairman Dr. Patrick Sensburg: That seems pretty complicated to me. Witness Klaus Landefeld: That’s always a question of whether the operator cooperates or has to cooperate, let’s say so, yes (Deutscher Bundestag, 2015, p. 16).

Indeed, alternatively, having a camouflage autonomous system just to connect to DE-CIX would not generate the results expected, as an IXP customer does not have access to any data indistinctly, and it would need to work as a common company, with customers and so on to attract any traffic. In Landefeld’s words: “It would certainly be more helpful to cooperate with any other company that is well known, I say, or that perhaps - for whatever reason - wants to cooperate with you” (Deutscher Bundestag, 2015, p. 51).

Secondly, in arguing about the difficulties of alternatively tapping their fibers in the city with malicious purposes, he says:

273

It may be so easy to imagine. But it’s actually extremely difficult to actually do that. This is only possible if you somehow have a piece of paper in your hand, which says: ‘I’m allowed to do that’, and then you actually have a chance to tap into a glass fiber with us in the Frankfurt area (Deutscher Bundestag, 2015, p. 25).

Two main messages stand out in Landefeld’s hearing. One is that of the consequent need of the state to extend its networks to private actors when it comes to surveillance through the internet interconnection infrastructure. In this sense, similar to what has been described in other policy sectors, the private sector coopera with the state and co-opera its surveillance mechanisms (Cardoso, 2018)—cooperates (collaborates), and co-operates (operates together)— playing a singular role in the surveillance assemblage.

The other is the key role of another actor to engender the public-private relationship: the legal apparatus. Below, once more, in answering about the possible access of other than German intelligence agencies to DE-CIX, the barrier that formally prevents foreign secret services from getting to DE-CIX is the same that allows BND to access it:

Chairman Dr. Patrick Sensburg: (…) Can you confirm that for DE-CIX: There was no contact, no contact with the NSA, the CIA [Central Intelligence Agency], GCHQ [Government Communications Headquarters] or other secret services of the Five Eyes States? Did they not go to DE-CIX with the question, though it’s nice to say, ‘Can we do something?’? Witness Klaus Landefeld: No. There was not even a contact attempt. Well, they did not approach us at all. And of course we would have rejected any negotiation so immediately. Well, there is no legal framework for it at all (Deutscher Bundestag, 2015, p. 21, emphases added).

What is explicit in Landefeld’s testimony is the prominence of the national interests materialized in the local law, and how this ends up being a substantial support to extend the state associations in order to conduct surveillance. As he admits: “[Y]ou have to deal with the laws of each country and must comply with them. It does not matter how” (Deutscher Bundestag, 2015, p. 45). Because an IXP involves a distributed network of actors too, including data centers,

274

telecommunications equipment providers, optical fiber carriers, etc., physically tapping an internet node without the IXP collaboration, although possible, is costly and is always under the risk of being uncovered at a certain point. The law becomes a great shortcut

What is absent in the previous discussion are the consequences for foreign citizens whose information is circulating through DE-CIX due to its prominence as a giant IXP worldwide. It is worthwhile to raise questions about the rights to privacy of personal information that, for instance, Brazilian individuals like me have. In 2014, in the midst of Edward Snowden revelations, the Brazilian Congress passed the Brazilian Internet Bill of Rights, originally Marco

Civil da Internet, which legally protects, among other provisions, the privacy of internet users in different instances (Rossini et al., 2015). With the increasing attraction of Brazilian ISPs to DE-

CIX, it is worth calling attention to the lack of legal protection to citizens’ privacy beyond

Brazilian borders, and moreover, the clear imbalance of power between a U.S. law like CALEA, which ends up being embedded in telecommunications equipment used worldwide, and that of a global South country as Brazil, which despite receiving global attention and being an international reference (Rossini et al., 2015; Santoro & Borges, 2017), has limited practical effects in the context of global internet communication. This is a fundamental open question that once more highlights major gulfs between the North and the South in regard to the infrastructure interdependencies and jurisdiction issues that pervade IXPs.

3.3.IXPs as “Centres of Appropriation” of a Surveillant Assemblage

IXPs were neither created for surveillance purposes nor do they enable a completely new surveillance state. They are just convenient spaces where a government can continue to 275

implement its long-standing monitoring strategies with the support of internet intermediary organizations. Nevertheless, IXPs’ affordances entitle states to extend the monitoring to new levels. Agents’ surveillance goals are delegated to IXPs’ cables, routers, switches, servers and datacenters facilities in a translation process (Latour, 1999) where a social phenomenon with humans and non-human actors arises. In this context, IXPs are “centres of appropriation”

(Haggerty & Ericson, 2003), which depend on a network of relations that may include laws, judicial orders, political agreements, and technological artifacts supporting altogether the collection of data to be further processed for matching surveillance goals.

Surveillance assemblages are multiple and modular, formed by many other assemblages.

They are also “emergent and unstable” (Haggerty & Ericson, 2003, p. 609), which means that they may change, and are not controlled by a single force. This is important for discouraging the focus on one institution as the heart of the problem—for instance, the BND in Germany, or any others that emerge. The desire of control is opportunistic and may be expressed through different channels, at different points of time.

The DE-CIX case shows that the implementation of surveillance mechanisms motivate state agents to learn about technical issues and take advantage of them for their own control purposes—which may have financial interests as well (Haggerty & Ericson, 2003). This implementation and motivation are not completely open, though. Klaus Landefeld complains that even if the procedures are lawful, one is not allowed to publicly talk about them—neither about the contact of an intelligence agency nor about the orders received. In his words, “This is actually the case in just about every country, that these orders are just then secret” (Deutscher

Bundestag, 2015, p. 45). Public access to a surveillant assemblage is extremely scarce. It is also

276

not completely clear how network operators are informed about the tapping of their lines when that happens through the IXP and not with their direct cooperation.

Importantly, public awareness about the case is not a result of DE-CIX procedures acting as a transparent organization. Instead, this information is publicly available due to whistleblowing and leaks that led to a formal hearing process. On the DE-CIX webpage, a release that reports that the IXP filed a lawsuit against the Federal Ministry of Interior in

Germany questioning the legality of the legal orders received from the Federal Intelligence

Service was published only in September 2016 (DE-CIX, 2016), seven years after the first conversations start. DE-CIX commercial operations does not seem to have been affected negatively. In an interview with an international network operator, the interviewee did not approach the surveillance case spontaneously, and expressed a position that it is inevitable. As long as it does not affect the quality of service as understood in engineering (low latency), that does not become an issue:

This is not an engineering concern that Germany has a regulatory framework that [allows] the state to intercept. Dot. That would not discourage me in any way [to interconnect in an IXP in Germany] because I do not see it as a barrier. [This will not change] (...) my guiding principles of peering. Improving latency is the main metric that makes me go to a traffic exchange point. Dot. Decrease my RTT [round-trip delay time]. Dot. (...) Within this, no regulatory framework related to privacy is relevant. You’re talking to someone looking at traffic engineering. My mission is to optimize. Now, if the regulatory framework for interception hampers traffic, that's another factor. But that does not happen.

Despite this position, if one considers on the one hand, the public interest function of internet exchange points, where an extraordinary amount of data from people beyond geographic borders traverses, and on the other hand, the requirement of transparency and accountability directed to organizations in charge of public goods in democratic contexts, the surveillance case at DE-CIX, the way that it came up to the public, and the very late position of DE-CIX about the 277

case are critical aspects of the internet infrastructure and contemporary communication that should not be normalized.

4. Law as a Boundary Object

A local intelligence agency, such as the BND, counts on controversial interpretations of the law to support its surveillance practices. In its turn, a foreign intelligence agency, as the

NSA, looks for partnerships to have access to more data than that which it is legally allowed to access within its own territory, offering its peers, in turn, enviable technology and know-how.61

Likewise, at the implementation level, legal order interpretations are also needed and legitimation from legal institutions may be sought. As Klaus Landefeld clearly exemplifies:

And those were points [in the beginning] where we said: ‘you have to discuss that.’ That was also at that time not at all clear how it is with the traffic. Can we then divert 100 percent of a line, even if there is something of 20 percent [rule allowed in the law] or something? At that point, we had no experience with how to handle packet-based communication in this case. And that was the reason we said we see a need for discussion here. And since, in our view, the licensing authority was the G-10 Commission, we wanted to talk about how they see it for the first time. - That was our procedure. (Deutscher Bundestag, 2015, p. 59).

Landefeld also remembers that after mirroring a line, it is not possible for the providers who are being intercepted to know what happens with that data, if its use is following the rules. In fact, the checks and balances are internal to the state.

On that, Landefeld’s impressions led him to conclude after government meetings that different state departments were not aligned about what was being done (Deutscher Bundestag,

61 This happens not without tensions, though. Partnerships in surveillance procedures mean to feed not only itself, but the ally too, sometimes against the interests of other national agencies and their own sovereign states, in a complex yet obscure web of relations that deserve further study. 278

2015). Similarly, revelations on the Operation Eikonal suggest that BND conducted actions without the awareness of the Chancellery, its superior department (Mascolo et al., 2014) . When it comes to collaborations with other states, the difficulty in monitoring compliance with the laws and the effective operation of checks and balances is still a concern. That was one of the reasons for early red flags raised by BND legal staffers when, along with NSA, the German agency was collaboratively conducting the Operation Eikonal. According to the news, these staffers

(…) wrote a memo explaining that the German Intelligence service, due to its ’technical inferiority’ could not verify whether the NSA did in fact comply with its promises, and that “full control” with the exchange of data, therefore “in reality was not possible.” As a consequence, BND warned about potential ‘political damage’ from the collaboration and “insurmountable consequences” (Geist et al., 2014).

Thus, under the rule of law, state massive surveillance develops amid flexible interpretations of legal instruments and undetermined situations creating an intricated scenario that leaves the surveillance black box semi-open, subject to intense debates.

Boundary objects are considered to be subject to “interpretive flexibility” (Star, 2010), and to express “work arrangements that are at once material and processual” (Star, 2010, p. 604).

This means that the social and material factors that surround boundary objects matter. Looking at legal documents as boundary objects contributes to the analysis in that it becomes clear that the mechanisms that have helped a surveillance state to extend its reach, including the collaboration of a business such as an IXP, are a product of negotiation at different levels, involving the acceptance of its terms but also the shape of the mechanics by the providers themselves, who have the know-how and the machinery to make the interceptions. This is a key characteristic of the co-operation of businesses in the state surveillance practices.

At the time of this writing, the DE-CIX lawsuit against the German state is under judicial secrecy, and professionals working with the case are not allowed to talk about it. 279

The litigation has substantial support. In an expert opinion piece, demanded by DE-CIX, about the reasonability of the German Federal Intelligence Agency actions at the Frankfurt IXP, the Former President of the Federal Constitutional Court, Hans-Jürgen Papier, concluded that

“Access by the Federal Intelligence Service to a data exchange point such as DE-CIX is unlawful on the whole” (Papier, 2016, p. 15, own translation).

Simon Assion, a German attorney member of the team responsible for the case, resonates such critique, recognizing in an interview with the author by e-mail that: “The BND has not been under a level of democratic control that would be necessary.” For him:

Quite often the [intelligence] agencies adopt ‘secret interpretations’ of the law, then act on the basis of their own interpretation, and no external control is exerted because the control bodies are never informed of this. The German System that should control and supervise intelligence agencies failed to detect and prevent this.

Despite substantial criticism, he recognizes that recent law reforms have given more power to the German Intelligence Agency: “The BND Act [Germany (2016)]is basically an attempt to legalize activities that have been going on for many years, but that were uncovered only recently.” Thus, even if there are reasons for DE-CIX to file a lawsuit trying to keep its level of reliability as a cross-border organization, the German political context will be a determinant of the outcomes of the litigation. Importantly, if laws are changed to increase the power of intelligence agencies, and court decisions favor surveillance acts, that will be considered legitimate for German citizens as these changes happen under democratic rules, even if those results may not be desirable for some. However, that is not legitimate for foreign citizens, though, who are not under the sovereignty of such a nation state.

On the other hand, from Landefeld’s hearing and public talks it is clear that while it is completely unacceptable that foreign intelligence agencies have access to the German data, as

280

previously seen, BND’s work seems to be acceptable to a certain extent, with regard to law enforcement functions conducted on a case-by-case basis, to fight crimes based on specific evidences. Massive surveillance, though, or the so-called U.S. NSA “collect it all” strategy and its diffusion around the globe is not well regarded. This is good evidence for the existence of a

“leaky” surveillance black box nowadays, which can be seen in the negotiations of the limits for the internet to become or not to become openly the largest surveillance apparatus of all times, and consequently, for IXPs and other intermediaries to become or not to become institutionalized

“centres of appropriation” for massive data collection. In a public talk this tension is made explicit:

So, the question must be: why do we accept this? Why is it acceptable that rights such as the confidentiality of communication, the inviolability of the home and similar should no longer apply [in the digital world], when they were so hard-fought in the analogue world? Why is the digital diary, the picture series on the net, the cloud storage, in short, the whole digital life we lead, subject to less protection than the analogues from it? The same applies, incidentally, to monitoring without a specific reason, i.e. monitoring on a case-by-case basis. The answer so far is simply to be able to do everything. There is no good reason. For the practical training, it is downright shocking to observe that the government reacts to the disclosure of Edward Snowden [with] more surveillance. Specifically, with the question: how can such possibilities [NSA activities] also be achieved? How can we catch up with the other states? That is extremely frustrating when you are being asked as a supplier. More data collection, an extension of monitoring also in our country, our own services: that is the answer of the legislature. The capacities of the German Federal Intelligence Agency [BND] and the Federal Office for the Protection of the Constitution [Verfassungsschutz] have increased by a factor of ten since 2013. (TEDx Talks, 2015).

This tension in values is materialized by a modus operandi of collecting and monetarizing data massively which characterizes the current surveillance capitalism pioneered by Google

(Zuboff, 2015, 2019), but which by no means is restricted to it. Interestingly, Landefeld calls

281

NSA “the Google of the secret services”62 (Deutscher Bundestag, 2015, p. 26), indicating how data has been greatly exploited in government as well.

The collection is massive even when the search is focused on selectors’ lists, with specific words of interest. As long as there is capacity to store data, state intelligence agencies and companies put extraordinary efforts into data mining and that seems to be the current standard and expected way to do it: collection of massive data; data treatment and action on behavior (Rouvroy & Berns, 2018). As Antoinette Rouvroy and Thomas Berns explain, in such a context, “‘false positives’ will never be interpreted as ‘failures,’ since the system follows a tracking logic rather than a diagnostic logic: the goal is not to miss any true positive, whatever the false positive rate is.” (Rouvroy & Berns, 2018, p. 117, own translation). From a massive surveillance apparatus vantagepoint, it is not a problem that would focus attention on a common researcher because he/she works with international drug traffic. As a scholar complained, along with journalists that cover such matter, they would be more likely to be surveilled after reforms in the G-10 Act in German, which has increasingly expanded the state monitoring of telecommunications capacities (Kommers & Miller, 2012, p. 414).

Importantly, massive surveillance is not separable from targeted surveillance—in fact, the latter follows the former—the targeting should not be dismissed from analysis.

In line with the attention that the researcher above is bringing to her/himself, let’s consider, for instance, the United States’ Department of Homeland Security (DHS) keywords used to monitor social media. Such selectors are distributed among categories of interest that include “DHS & Other [Government] Agencies;” “Domestic Security;” “HAZMAT & Nuclear;”

62 “die Google der Geheimdienste wäre” 282

“Health concerns + H1N1;” “Infrastructure Security;” “Southwest Border Violence;”

“Terrorism;” “Weather/ Disaster/ Emergency;” and “Cybersecurity” (U. S. Department of

Homeland Security, 2011, pp. 20-23). Interestingly, among the keywords that appear in the list are places such as El Paso (Texas), Mexico and Sonora (Mexico)—that could lead any of my friends from these places publishing about their home on social media to have their posts monitored. And although the DHS guidelines are clear about the need for its staffers do not include “Personal Identifiable Information” in their reports unless before “extremis situations” involving threats of death, damage of critical resources, etc. (U. S. Department of Homeland

Security, 2011, p. 24)—a fact that could keep my friends’ identity safe—the truth is that identifiable information is likely to be part of data collected, requiring an additional effort to be ignored.

States’ laws, interpretation of the laws, selector lists and keywords permit access to the explicit targets. These targets are individuals with particularities. Thus, when data collection is conducted at IXPs, what data means needs to be problematized. As authors have argued, “a piece of data is nothing more than a sign expunged from all of its own signification” (Rouvroy &

Berns, 2018, p. 112, own translation). Data packets are then individuals’ pieces; parts of bodies virtually traversing cables, routers, switches and geographic located nodes in internet infrastructure. And as bodies have never been equal in society, there is no reason to consider that they will be when they are digitalized and have their characteristics inspected. Laws are important elements to define these differences, and which bodies are more likely and acceptable to be inspected. They are “boundary objects with agency” (Fleischmann, 2006) in this regard.

283

5. The Embodiment of Data Packets

In Germany, foreigners’ information has recognizably received fewer legal protections than national citizens. Former G-10 Commission chair Hans de With explained in his hearing that the BND has the authority to monitor international communication traffic and for that it is not necessary for any control agency approval (Deutscher Bundestag, 2015). Additionally, in his expert opinion piece, Hans-Jürgen Papier also confirms that, according to the G-10 Privacy Act, also known as the BND Act, “the Federal Intelligence Service may (…) strategically monitor international telecommunications traffic with specific countries or regions in addition to targeted data collection” (Papier, 2016, p. 2, own translation). Simon Assion critiques this legal situation:

The BND takes the position that—despite clear previous ruling of the Constitutional Court—persons outside of Germany have no right to communications secrecy (so-called “foreigners theory”). This interpretation has no basis in the Constitution. (…) This is why many observers were very upset that the BND had secretly ignored this decision for many years. Many observers saw this as a challenge to the rule of law itself, which is something very serious when coming from an intelligence service.

Beyond targets’ nationality differentiation in the way that surveillance practices have taken place, the Bundesdatenschutzgesetz (also known as BDSG) (Federal Data Protection)

(Germany, 2009), since 2003 defines “special categories of personal data” that are not allowed to be collected and processed by default. That includes: “information on a person’s racial or ethnic origin, political opinions, religious or philosophical convictions, union membership, health or sex life” (SECTION 3, (9). p. 3). Exceptions for this rule occur if “5. such collection is necessary in order to avert a substantial threat to public safety, 6. such collection is necessary in order to avert substantial detriment to the common weal or to protect substantial interests of the common weal” (SECTION 13, p. 14), among other situations.

284

Recently, with the emergence of new European laws, such as the General Data Protection

Regulation (GDPR), the German Federal Data Protection (Germany, 2017, official translation) has become more comprehensive in regard to collection and processing of “special categories of personal data,” which include:

a) data revealing racial or ethnic origin, political opinions, religious or philosophical beliefs, or trade union membership; b) genetic data; c) biometric data for the purpose of uniquely identifying a natural person; d) data concerning health; and e) data concerning a natural person’s sex life or sexual orientation; (SECTION 46, 9, p. 25-26)

The national legislation now refers to GDPR in regard to the exceptions to processing such data. For instance, it is not prohibited if:

(f) processing is necessary for the establishment, exercise or defense of legal claims or whenever courts are acting in their judicial capacity; (g) processing is necessary for reasons of substantial public interest, on the basis of Union or Member State law which shall be proportionate to the aim pursued, respect the essence of the right to data protection and provide for suitable and specific measures to safeguard the fundamental rights and the interests of the data subject; (ARTICLE 9(2), p. 38)

With that, it is important to highlight how such categories ultimately compel the embodiment of data packets containing personal data in a surveillance context. In other words, it makes the case that data packets are not detached from an individual’s body and identity.

One more example comes from Hans de With’s hearing, where the Parliamentary member Andrea Lindholz included in her inquiry an excerpt from the news piece that first revealed the existence of Operation Eikonal. The paragraph reads as follows:

In order to get to transit traffic, a rather creative and even in the view of some BND experts inadmissible interpretation of the G-10 law should help: “G-10 is a door opener for the detection of international traffic,” noted a BND official, and exactly promptly protested a sub-department head against this approach. That would be unlawful, and transit traffic - at least if they were intercepted by means of a G-10 permit - should be searched only for terrorists or arms dealers under the narrow conditions of this law. But how can you delete trace when the data is passed on? His department head weighed: Everything was fine, everything covered by the BND Act (Deutscher Bundestag, 2015, p. 111; Mascolo et al., 2014, emphases added). 285

Similar to what occurs in airports, where surveillance procedures with X-ray machines have general scope but, at the same time, have a focus toward certain racialized bodies—the imaginary terrorist characters—, the so-called “massive” surveillance, specifically the one that starts collecting data through internet infrastructure, has the capacity to be diffused and targeted at the same time on some individuals or imagined individuals. As Rachel Hall shows with the study of scanners at the airports, there is the increasing strategy of building “compulsory transparency” to differentiate gentle citizens’ transparent bodies from terrorists’ opaque bodies

(Hall, 2015, p. 128), transparency.” For the author, “Like whiteness or heterosexuality, transparency claims the ground of neutrality, while in fact the transparent body desired by the security state is not neutral but, more accurately, normate” (Hall, 2015, p. 133).

Importantly, in the context of the digital era, information theories have built vastly on the liberal subject complained by critical feminist authors. As Katherine Hayles shows us, the most influential understandings of “information” that were raised in the 1950s with Alain Turing’s test to prove if a machine could think, consolidated the idea that information is something dematerialized and disembodied. Once detached from the body, information loses its matter and is seen as something that both humans and machines can process almost indistinctly. Ultimately, the body does not add to the information of the self.

Hayles wisely explains that this distance between information and matter, where the latter becomes secondary, feeds the basis of cybernetics theories. Further, it influences the emergence of a posthuman subject, who through the dissociation of body and self, can be understood as “a material-informational entity whose boundaries undergo continuous construction and reconstruction” (Hayles, 1999, p. 3). This is a key argument to bring context to the internet,

286

where the prevalent imagery is to see data packets as dematerialized entities and components of disembodied subjects that move through a distributed arrangement built on an unrealistic “free flow” of information. Under this paradigm, message content, including characteristics of origin and destination, are subsumed, while the universality of the liberal political theory subject prevails. This scenario averts discussions on the materiality of the data packets; the (non- universal) subjects that they are representing; the media through which they pass; and the interests that are raised by these silenced characteristics in different points of the network for opportunistic actors to collect data following specific interests materialized into keywords.

Data packets encapsulate our subjectivity and carry identifiers by default, allowing the tracing of origin, destination, and other sender’s and receiver’s attributes.63 Data packets traversing IXPs are thus impossibly disembodied: they correspond to individuals who have certain nationalities, political positions, religions etc., and opportunistic actors conducting data packet inspection in search of these bodies see them as so. Given the procedures of what authors call “algorithmic governmentality” that prioritize correlations to the detriment of context

(Rouvroy & Berns, 2018, own translation), algorithms can easily correlate identifiable characteristics to the opaque bodies of wanted terrorists—e.g. information collected with identifiers from certain regions, with certain words, may be flagged as suspect, fueling the surveillance assemblage operations, with already studied consequences.

63 Despite the focus on the subject here in this chapter, I would like to recognize that objects characteristics are also encapsulated into data packets when digitalized, making the digital embodiment a process that applies to humans and non-humans. A good example is the song “Pela Internet” (Through the Internet), from the Grammy Award winning album “Quanta” (1998) by Gilberto Gil. The lyric says: “With how many gigabytes/ One makes a raft, a boat that sails/ That sails in this infosea? (own translation). I am grateful to Maria Teresa Rocco who remembered me about this song. 287

Characteristic of the surveillance capitalism, the “Big Other” is considered a ubiquitous mechanism that “aspires to encompass and reveal the comprehensive immanent facts of market, social, physical, and biological behaviors” (Zuboff, 2015, p. 82). In other words, everything that is possible based on any trace that exists, and from which it is not possible to hide. It generates a specific kind of behavior that is absorbed and becomes habitual, as the concept of “anticipatory conformity” elucidates:

Anticipatory conformity assumes a point of origin in consciousness from which a choice is made to conform for the purposes of evasion of sanctions and social camouflage. It also implies a difference, or at least the possibility of a difference, between the behavior one would have performed and the behavior one chooses to perform as an instrumental solution to invasive power (Zuboff, 2015, p. 82).

Such effects have other sensitive dimensions associated with what Julie F. Cohen has called the configuration of the self. Specifically, she states that “networked information technologies differ from other technologies in the extent to which the technological mediation of embodied perception is not simply functional, but also representational. The effects extend to our own self-perception” (Cohen, 2012, p. 49, emphasis added). She defends the importance of not dismissing the body from the analysis—a task historically taken seriously by feminist scholars.

Feminist scholarship has criticized approaches that look at surveillance exclusively in terms of privacy and under the principles of the liberal political theory, where the liberal subject is assumed to be universal—silencing important identity markers. For these scholars—who emphasize both the historical deficit of privacy to some social groups, such as immigrants, prisoners, the non-educated, women and especially women of color, to name a few, and the ways that violence has affected women at the privacy of the home—, it is necessary to reconcile privacy with other rights, taking different identities into account (Bhattacharjee, 2001; Gilliom,

288

2001; Magnet, 2011; Roberts, 1991). They bring strong evidence of how the body is an essential category to surveillance studies.

The concept of “incodification” implies the existence of the body and conveys the idea that subjects are trapped “within a web of legal, bureaucratic and digital codes” (Lingel &

Sinnreich, 2016). Jessa Lingel and Aram Sinnreich build the concept of incodification from the phenomenon of institutionalized incarceration in which social coercions permeate people’ lives regardless of being physically incarcerated or not. The logical question that comes from the comparison between institutionalized incarceration and digital massive surveillance is the fact that if institutional incarceration affects certain social groups more than others—black people, immigrants, the poor, etc.—, how is incodification any different to imagine it as indistinctly massive? What are the keywords that state agents use to sift data if not the materialization of the bodies that are red flagged and potentially wanted? Considering this context, data collection at

IXPs becomes the beginning of a long chain of a surveillance assemblage waiting to be further uncovered.

6. Final Considerations

State surveillance in the digital era involves state and corporate actors working together.

The way that the state builds associations with private actors varies and involves money, law, and the consequences of that. Legal documents are far from being objective and undeniable, generating disputes around interpretations, as well as law reforms to renegotiate the limits of actors’ power. The technical expertise of corporate actors shapes the way that surveillance is conducted and introduces a tension. In the case studied here, the tension is established between 289

the German state desire to grow its surveillance capacities and the will of a company such as DE-

CIX to be a multinational and a leading IXP in compliance with good practices in engineering but also the legal environment.

As a company, DE-CIX opts for keeping the information hidden but it becomes public when Snowden’s leaks make it inevitable. Although the tensions and debates show that the state surveillance black box is still open and the terms of its functioning can alter, it is clear that commercial interests and the apparent distant connection between IXPs and end users affect the way that the public interest is understood and taken into consideration. If other voices were closely heard, as the ones from the demonstrations led by civil society as Amnesty International in Berlin asking for the “stop of German NSA” (Amnesty International, 2016), would DE-CIX posture be the same?

This is when multistakeholder bodies gain relevance. As IXPs are key infrastructure intermediaries in our virtual transportation from one point to another on the internet, they need to be constantly reminded of their role and their work in the public interest. A multistakeholder governance—not only with its direct customers as it currently exists, but with digital rights representatives, scholars and end users’ associations—is a way to do so and embed societal values into the design of IXPs. The key challenge for civil society is to show how this can bring

(commercial) benefits to such IXP organizations.

To understand massive surveillance in the context of the internet, information and potential information—as data packets are—need to be reembodied, as Katherine Hayles suggested almost two decades ago. It is not difficult to foresee that such an argument may be considered difficult to integrate in policy discussions where concepts such as net neutrality are

290

intended to make Internet Service Providers disregard any identifying characteristics of data packets, which may (undesirably) be used to prioritize certain traffic motivated by market purposes. However, embodying information by no means intends to be an argument against the need to maintain ISP agnosticism. On the contrary, it may help to leverage ISPs’ agnostic role, by deeply understanding what they have ignored in data packets traversing their networks, which opportunistic actors have not. This understanding may help other civil society actors to be more deeply involved in interconnection matters, creating standards and guidelines that, as norms, can help better regulate this yet opaque field.

As laws such as GDPR evolve, the demand for information embodiment to protect the rights of their citizens will likely become more present—e.g. a possible question to answer in order to know if GDPR has been followed is where European Union citizens’ data is circulating.

The discussions on the technical ways to answer this kind of question will certainly advance public understanding of how internet data traffic and data packets inspections currently occur.

291

CHAPTER 7

CONCLUSIONS

Nothing is so oppressive as to eliminate the sense of a nonoppressive alternative. But, on the other hand, no such alternative is strong or convincing enough to avoid running the risk of somehow conflating itself with oppression.

—Boaventura de Sousa Santos, Epistemologies of the South (p. 12)

On October 18, 2017 at 5:30 pm I was at the Frankfurt Airport for a “Ready for take-off” tour offered to the attendees of the annual DE-CIX Summit that would take place the day after.

We were around thirty to forty people, mostly DE-CIX customers’ representatives—embodied networks with whom I could talk and enjoy the afternoon. The tour was great. On a bus, we were taken to the airport runway, aircraft parking areas and were given explanations about the airplanes machinery’s details, models and capacities, and the curiosities of the largest airport in

Europe—sometimes considered the second after the Charles de Gaulle in Paris, the guide would alert. The time was perfectly chosen in that we could see the beautiful sunset at the end of the tour.

I said to a DE-CIX staffer that it would be amazing if a similar tour could be done through DE-CIX data centers dependencies as well, and was told that this would not be that interesting to their customers, as they are very used to that technical environment. I had to agree.

That airport simulations about people transportation were truly a different way of experiencing data transportation, and indirectly, global communication via the internet.

292

The day after also followed the same line. The summit took place in a charming town, with traditional German architecture, connected to Frankfurt by train. The venue simulated an airplane, the seats, the windows, the corridor, all was surprising. I was fortunate to be there, having an experience in which an IXP was simulating that we, the people, were all in transit, being transported as passengers (or packets?) to somewhere yet unknown.

I had arrived there traveling from Brazil. An ISP owner near me in the airport hall came from Italy. Meeting in Germany was not a coincidence, but the result of our common, yet distinct, interest in the largest IXP in the world. I was of course interested in understanding the reasoning for the commercial agreements that justified such an event and its attendance, but my ultimate concern was how that connected to my motivation as a researcher, adding to the understanding of the sociotechnical mediation of our lives. An answer that I can finally have now.

In 2005, at the World Summit on Information Society in Tunis, Stefano Rodotà, a well- known Italian jurist, and Gilberto Gil, a famous Brazilian singer and then Ministry of Culture, were the first two signers of a letter defending the construction of an Internet Bill of Rights based on principles to guarantee a “nuova cittadinanza planetaria” (new planetary citizenship)

(Key4biz, 2005). That evolved to be a workshop of a “Dynamic Coalition for the Internet Bill of

Rights” in the first Internet Governance Forum in 2006 in Athens, and then a “Joint Declaration on Internet Rights” between Brazil and Italy during the Internet Governance Forum in 2007 in

Rio. The proposal envisioned continuing discussions on the topic and pointed to the importance, among other issues, of the “global reachability of all Internet nodes” (Gil & Vimercati, 2007).

293

Discussion on internet principles has generated numerous local initiatives and outcomes

(NETmundial, 2014a). The Brazilian Internet Bill of Rights, passed in 2014, is an example of a pioneering law which has inspired discussions in other countries, Italy included. This discussion gained world proportions again during the NETmundial, a multistakeholder event that convened more than 1,400 people from 97 countries in São Paulo and remotely (NETmundial, 2014b), soon after the Snowden revelations in 2013. There, a non-binding declaration with global internet principles was issued as the result of 180 contributions from different countries and stakeholders (NETmundial, 2014c).

These efforts altogether call for the recognition that a global communication medium needs to be founded on principles that respond to values that transcend one nation. That while privatized, the internet infrastructure needs to attend to the public interest, and that public is transnational.

Based on the present research findings, general principles to guide internet exchange points worldwide are necessary. These internet infrastructure nodes are pivotal in the interplay between local and global spheres, subject to highly controversial exploitation by states that follow questionable interests, and not uncommonly, relegate fewer rights to foreigners than to national citizens. Principles guiding IXPs can take advantage of previous efforts proposing internet principles founded on human rights related to freedom of speech, privacy, and others that have been discussed since the 2000s64, along those values of users and non-users, from urban and rural areas, and of the congregation of IXPs and their associations worldwide. While common people can share practical situations that need to be addressed on the ground, IXP

64 An initial comprehensive list is available at NETmundial (2014a). 294

operators are able to bring the main threats to the guarantee of engineering good practices and the so-called and valued network operators’ agnosticism at the level of internet interconnection infrastructure. The conversation between values and rights languages and engineering codes may need mediation, but nonetheless has the potential to generate unprecedented positive results to help IXPs operators to have support in responding to states controversial onrushes.

A global discussion on the IXP ecosystem with an economic lens and further code ethnography studies are also necessary to contribute to balance the negative consequences of a one-way flow of information going from the South to the North. Definitely, business concentration needs to be problematized at the level of internet infrastructure and not only when it refers to content providers. The discussion is complex because in national markets there is no traditional way of measuring competition among IXPs, and additionally, some of them are not for-profit and do not compete per se among each other. Nevertheless, the problem of concentration is seen when comparing IXPs’ performances worldwide—an effort that the present dissertation does, revealing the highly different level of connectivity that a node in the North allows autonomous systems to have. The scenario points to a continuation of these concentration patterns as the more connected networks at a node, the more new networks tend to come. But considering the global South and global North symmetrically, such concentration is highly problematic.

As elucidated by the present dissertation, the IXP ecosystem is marked by the late development of IXPs in the global South and by the existence of giant IXPs in the global North that benefit from both their common interests with big content delivery networks and the lack of infrastructural resources in countries from the South. That is observable by shedding light on the

295

consistent concealment of the South in the study of internet infrastructure. Illuminating that concealment leverages our public understanding of internet interconnection politics and rebuts the paradigm of “free flow” of information.

Another point that is urgent to address are the values that arise from the discussion of internet interconnection included in the absence of IXPs. This research has shown how the very moment of network interconnection in rural areas becomes key to understanding the role of first

(also known as last) mile internet infrastructure to challenge and co-produce societal values among Tseltal and Zapoteco communities. It has also addressed the disparities and the lost link between the need for internet of these indigenous communities and the deployment of a distant

IXP covered with the narrative of potential social benefits but embedded in commercial interests and subject to the market power of an incumbent.

By leveraging public understanding of internet materiality, raising attention to societal values, this dissertation joins scholarship that is calling for more participative fora where common people can take part in the design of aspects and artifacts that play important roles in their lives (Appadurai, 2013; Callon et al., 2011). Their infrastructuring activities not only deserve to be recognized as part of internet codesign, but their voices and experiences can lead to the development of other solutions and possibilities and should not be kept unheard, including in

IXP governance.

This dissertation has approached the numerous decisions and values embedded in people’s communication mediated by internet exchange points. Interestingly, as Mark Poster— concerned with the then new “electronically mediated communication” (Poster, 1990)—warned, the relationship between “words” and “things” has been impaired with the advent of computing

296

by the disappearance of the “referent,” to which the language is supposed to correspond. A good complement to this thought is the idea that technology “seeks to be forgotten” (Latour, 2013, p.

217), and that happens in part because society approaches technology as “means toward ends”: if they work effectively, the result is memorized but the process is left behind (Latour, 2013, pp.

217-219).

Communication that once corresponded to individuals speaking in person and expanded through other technologies such as letters, telegraph or telephone, now corresponds mostly to a multifaceted ecosystem in which data packets become an extension of the emitter’s body along with the information circulation infrastructure of the internet. The opacity of the procedures that occur in the underlying layers of internet infrastructure leads to a sophisticated process of representation and simplification that society needs to develop for the word communication to be meaningful, even with highly complex technologies involved. The general focus on means and ends—e.g. technologies seen as means to “connecting people,” a common slogan of communication corporations—, helps keep the process opaque.

It is fundamental to remember that communication, although keeping its ends, refers nowadays to a different “thing,” and includes several mediators and decisions of third parties not even known by emitter and recipient, but that should be under their scrutiny in order to be responsive to public values. A first step to do that is strengthening the bonds between the referent and the language when it comes to people’s communication. Broadening our repertory understanding of governance by IXPs, their presences, absences and significance, is the contribution of this dissertation towards this purpose.

297

REFERENCES

Abbate, J. (2010). Privatizing the Internet: Competing Visions and Chaotic Events, 1987-1995. IEEE Annals of the History of Computing, 32(1), 10–22. https://doi.org/10.1109/MAHC.2010.24

Abbate, J. (1999). Inventing the Internet. Cambridge, Mass.: The MIT Press.

Abranet. (2018, October 17). Anatel: Pequenos provedores detém 22% do mercado de banda larga. Retrieved April 12, 2019, from http://www.abranet.org.br/Noticias/Anatel%3A- Pequenos-provedores-detem-22%25-do-mercado-de-banda-larga- 2132.html?UserActiveTemplate=site#.XK5UlLjQ82w

Acker, A. (2014). The Short Message Service: Standards, infrastructure and innovation. Telematics and Informatics, 31(4), 559–568. https://doi.org/10.1016/j.tele.2014.01.004

Acker, A. (2015). Toward a Hermeneutics of Data. IEEE Annals of the History of Computing, 37(3), 70–75. https://doi.org/10.1109/MAHC.2015.68

AFIX. (n.d.). List of active Internet exchange points in Africa. Retrieved April 7, 2019, from https://www.af-ix.net/ixps-list

Afonso, C. A. (2004, September 21). O complicado mundo da ICANN. NIC.br. Retrieved from https://cgi.br/publicacao/o-complicado-mundo-da-icann/

Africa Union. (n.d.). African Internet Exchange System (AXIS) Project Overview | African Union. Retrieved April 17, 2019, from https://au.int/en/african-internet-exchange-system- axis-project-overview

Ager, B., Chatzis, N., Feldmann, A., Sarrar, N., Uhlig, S., & Willinger, W. (2012). Anatomy of a Large European IXP (pp. 163–174). Presented at the SIGCOMM’12, Helsinki, Finland.

Agudelo, M., Katz, R., Flores-Roux, E., Botero, D., Cristina, M., Callorda, F., & Berry, T. (2014). Expansión de infraestructura regional para la interconexión de tráfico de internet en América Latina. CAF. Retrieved from http://scioteca.caf.com/handle/123456789/522

298

Akamai. (2018, April). Real-Time Internet Monitor | Akamai. Retrieved November 30, 2018, from https://www.akamai.com/us/en/solutions/intelligent-platform/visualizing- akamai/real-time-web-monitor.jsp

Almeida, V. A. F., Goh, B., & Doneda, D. (2017). A Principles-Based Approach to Govern the IoT Ecosystem. IEEE Internet Computing, 21(4), 78–81. https://doi.org/10.1109/MIC.2017.2911433

Almeida, V., Getschko, D., & Afonso, C. (2015). The Origin and Evolution of Multistakeholder Models. IEEE Internet Computing, 19(1), 74–79. https://doi.org/10.1109/MIC.2015.15

Álvarez, C. L. (2018). Telecomunicaciones y Radiodifusión en Mexico. Mexico City: UNAM Posgrado Derecho. Retrieved from http://www.dijuris.com/es/libro/telecomunicaciones- y-radiodifusion-en-mexico_30669

Amnesty International. (2016, October 21). Protestaktion in Berlin: Stoppt die “deutsche NSA”! Retrieved May 30, 2019, from https://www.amnesty.de/2016/10/21/protestaktion-berlin- stoppt-die-deutsche-nsa

Anderson, W. (2012). Asia as Method in Science and Technology Studies. East Asian Science, Technology and Society, 6(4), 445–451. https://doi.org/10.1215/18752160-1572849

Anderson, Warwick. (2017). Postcolonial Specters of STS. East Asian Science, Technology and Society, 11(2), 229–233. https://doi.org/10.1215/18752160-3828937

Angola Cables, & DE-CIX. (2018, May 8). Press Release: Angola Cables to become new DE- CIX reseller. Retrieved from https://www.de- cix.net/Files/190e1bb0ea8141f38db054d174ef0572cce42028/Angola-Cables-new-DE- CIX-reseller.pdf

Anzaldúa, G. (2012). Borderlands / La Frontera: The New Mestiza (4 edition). San Francisco: Aunt Lute Books.

Appadurai, A. (2013). The Future as Cultural Fact: Essays on the Global Condition (1 edition). London: Verso.

299

Arkko, J. (2019a, January 11). Distribution of Authors per Companies. Retrieved January 14, 2019, from https://www.arkko.com/tools/allstats/companydistr.html

Arkko, J. (2019b, January 11). Distribution of Authors per Country. Retrieved January 14, 2019, from https://www.arkko.com/tools/allstats/countryeudistr.html

Arriscado, J. N. (2009). O Resgate da Epistemologia. In B. de S. Santos & M. P. Meneses (Eds.), Epistemologias do Sul (1st ed., pp. 215–242). Coimbra: Edições Almedina.

Arthur, B. W. (2009). The Nature of Technology: What It Is and How It Evolves (Reprint edition). New York: Free Press.

AT&T. (n.d.). Evolution of the AT&T and BellSouth Brands. AT&T. Retrieved from https://www.att.com/Common/attrev1/331012_timeline_evolution16.pdf

Baca-Feldman, C. F., Huerta, E., Álvarez, M., Parra, D., & Velasco, K. (2018). Tejiendo autonomía tecnológica en los pueblos indígenas: telefonía celular comunitaria en Oaxaca, México. In L. Belli & O. Cavalli (Eds.), Gobernanza y regulaciones de Internet en América Latina: Análisis sobre infraestructura, privacidad, ciberseguridad y evoluciones tecnológicas en honor de los diez años de la South School on Internet Governance (1st ed., pp. 289–304). Rio de Janeiro, Brazil: Escola de Direito do Rio de Janeiro da Fundação Getulio Vargas.

Baker, F. (1995, June). RFC 1812 - Requirements for IP Version 4 Routers. Internet Engineering Task Force (IEFT). Retrieved from https://tools.ietf.org/html/rfc1812

Baker, F., Foster, B., & Sharp, C. (2004, October). RFC 3924 - Cisco Architecture for Lawful Intercept in IP Networks. Retrieved from https://tools.ietf.org/html/rfc3924#ref-14

Balding, C., & Clarke, D. C. (2019). Who Owns Huawei? (SSRN Scholarly Paper No. ID 3372669). Rochester, NY: Social Science Research Network. Retrieved from https://papers.ssrn.com/abstract=3372669

Bamford, J. (2009). The Shadow Factory: The NSA from 9/11 to the Eavesdropping on America. New York: Anchor.

300

Bardzell, S. (2010). Feminist HCI: Taking Stock and Outlining an Agenda for Design. In Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (pp. 1301–1310). New York, NY, USA: ACM. https://doi.org/10.1145/1753326.1753521

Bates, T., Smith, P., & Huston, G. (2019). CIDR Report. Retrieved from https://www.cidr- report.org/as2.0/

Baumgärtner, M., Blome, N., Gude, H., Rosenbach, M., Schindler, J., & Schmid, F. (2015, April 24). Spying Close to Home: German Intelligence Under Fire for NSA Cooperation. Spiegel Online. Retrieved from https://www.spiegel.de/international/germany/german- intelligence-agency-bnd-under-fire-for-nsa-cooperation-a-1030593.html

Belli, L. (2018). Las redes comunitarias y el principio de autodeterminación de red. In L. Belli & O. Cavalli (Eds.), Gobernanza y regulaciones de Internet en América Latina: Análisis sobre infraestructura, privacidad, ciberseguridad y evoluciones tecnológicas en honor de los diez años de la South School on Internet Governance (1st ed., pp. 205–233). Rio de Janeiro, Brazil: Escola de Direito do Rio de Janeiro da Fundação Getulio Vargas.

Bendrath, R., & Mueller, M. (2011). The end of the net as we know it? Deep packet inspection and internet governance. New Media & Society, 13(7), 1142–1160. https://doi.org/10.1177/1461444811398031

Berker, T. (2011). Michel Callon, Pierre Lascoumes and Yannick Barthe, Acting in an Uncertain World: An Essay on Technical Democracy: The MIT Press, Cambridge, London, 2009. Minerva, 49(4), 509–511. https://doi.org/10.1007/s11024-011-9186-y

Bhambra, G. K. (2014). Postcolonial and decolonial dialogues. Postcolonial Studies, 17(2), 115– 121. https://doi.org/10.1080/13688790.2014.966414

Bhattacharjee, A. (2001). Private Fists and Public Force: Race, Gender, and Surveillance. In Policing the National Body, The Color of Violence: Violence Against Women of Color. Boston: South End Press. Retrieved from https://www.academia.edu/421186/Private_Fists_and_Public_Force_Race_Gender_and_ Surveillance

301

Biermann, V. K. (2014, December 4). Bundesnachrichtendienst: Daten abfischen mit Lizenz aus dem Kanzleramt. Die Zeit. Retrieved from https://www.zeit.de/politik/deutschland/2014- 12/bnd-kanzleramt-eikonal-nsa

Bijker, W. E., Hughes, T. P., & Pinch, T. J. (1987a). Common Themes in Sociological and Historical Studies of Technology: Introduction. In W. E. Bijker, T. P. Hughes, & T. J. Pinch (Eds.), The Social Construction of Technological Systems: New Directions in the Sociology and History of Technology (pp. 9–16). Cambridge, Massachusetts; London, England.

Bijker, W. E., Hughes, T. P., & Pinch, T. J. (1987b). The Social Construction of Facts and Artifacts: Or How the Sociology of Science and the Sociology of Technology Might Benefit Each Other. In W. E. Bijker, T. P. Hughes, & T. J. Pinch (Eds.), The Social Construction of Technological Systems: New Directions in the Sociology and History of Technology (pp. 17–50). Cambridge, Massachusetts; London, England: MIT Press.

Blackman, C., & Srivastava, L. (Eds.). (2011). Telecommunications Regulation Handbook (Tenth Anniversary Edition). Washington, DC.: The International Bank for Reconstruction and Development / The World Bank, infoDev, and The International Telecommunication Union.

Bloom, P. (2015). La Telefonía Celular Comunitaria como Alternativa al Sistema Hegemónico de las Comunicaciones en México: Un estudio de caso de las nuevas iniciativas de la Sierra Juárez de Oaxaca (Masters in Rural Development). Universidad Autónoma Metropolitana, Mexico City.

Borning, A., & Muller, M. (2012). Next steps for value sensitive design. In Proceedings of the 2012 ACM annual conference on Human Factors in Computing Systems - CHI ’12 (pp. 1–10). Austin, Texas, USA: ACM Press. https://doi.org/10.1145/2207676.2208560

Bossewitch, J., & Sinnreich, A. (2013). The end of forgetting: Strategic agency beyond the panopticon. New Media & Society, 15(2), 224–242. https://doi.org/10.1177/1461444812451565

302

Böttger, T., Cuadrado, F., Tyson, G., Castro, I., & Uhlig, S. (2016). Open Connect Everywhere: A Glimpse at the Internet Ecosystem through the Lens of the Netflix CDN. ACM SIGCOMM Computer Communications Review (CCR, 1(48), 1–14.

Bowker, G. C., & Star, S. L. (2000). Sorting Things Out: Classification and Its Consequences (Revised edition). Cambridge, Massachusetts London, England: The MIT Press.

Braden, R. (1989, October). RFC 1122 - Requirements for Internet Hosts - Communication Layers. Retrieved from https://tools.ietf.org/html/rfc1122

Braman, S. (2012a). Internationalization of the Internet by design: The first decade. Global Media and Communication, 8(1), 27–45. https://doi.org/10.1177/1742766511434731

Braman, S. (2012b). Privacy by design: Networked computing, 1969–1979. New Media & Society, 14(5), 798–814. https://doi.org/10.1177/1461444811426741

Brasil. (1995, May 31). Portaria no 148, de 31 de maio de 1995 (Norma 004/95). Ministério das Comunicações. Retrieved from http://www.anatel.gov.br/legislacao/normas-do-mc/78- portaria-148

Brasil. (1997, July 16). LEI No 9.472, DE 16 DE JULHO DE 1997. Presidência da República. Retrieved from http://www.planalto.gov.br/ccivil_03/leis/L9472.htm

Brasil. (2003, September 3). Decreto No 4.829, de 3 de setembro de 2003. Presidência da República. Retrieved from https://www.cgi.br/pagina/decretos/108

Brey, P. (2000). Method in computer ethics: Towards a multi-level interdisciplinary approach. Ethics and Information Technology; Dordrecht, 2(2), 125–129.

Brito, S. H. B., Santos, M. A. S., Fontes, R. dos R., Perez, D. A. L., Silva, H. D. L. da, & Rothenberg, C. R. E. (2016). An Analysis of the Largest National Ecosystem of Public Internet eXchange Points: The Case of Brazil. Journal of Communication and Information Systems, 31(1). https://doi.org/10.14209/jcis.2016.23

Brock, G. W. (1995, April). The Economics of Interconnection. TCG - Teleport Communications Group. TCG - Teleport Communications Group. Retrieved from http://www.citi.columbia.edu/tcg/economicsinterconnection.pdf 303

Callon, M. (1984). Some Elements of a Sociology of Translation: Domestication of the Scallops and the Fishermen of St Brieuc Bay. The Sociological Review, 32(1_suppl), 196–233. https://doi.org/10.1111/j.1467-954X.1984.tb00113.x

Callon, M., Lascoumes, P., & Barthe, Y. (2011). Acting in an Uncertain World: An Essay on Technical Democracy. (G. Burchell, Trans.). Cambridge, Massachusetts; London, England: The MIT Press.

Callon, M., & Latour, B. (1981). Unscrewing the big Leviathan: how actors macro-structure reality and how sociologists help them to do so. In K. K. Cetina & A. V. Cicourel (Eds.), Advances in Social Theory and Methodology (RLE Social Theory): Toward an Integration of Micro- and Macro-Sociologies (1st Edition). London: Routledge.

Camilo Cardona Restrepo, J., & Stanojevic, R. (2012). IXP traffic: A macroscopic view. In Proceedings of the 7th Latin American Networking Conference, LANC 2012 (pp. 1–8). https://doi.org/10.1145/2382016.2382018

Cardoso, B. (2018). Estado, tecnologias de segurança e normatividade neoliberal. In F. Bruno, B. Cardoso, M. Kanashiro, L. Guilhon, & L. Melgaço (Eds.), Tecnopolíticas da Vigilância. Perspectivas da Margem (1st ed., pp. 91–105). São Paulo, SP: Boitempo.

Cardoso, F. H., & Faletto, E. (1979). Dependency and Development in Latin America (1st ed.). Berkeley: University of California Press. Retrieved from https://www.ucpress.edu/book/9780520035270/dependency-and-development-in-latin- america

Carey, J. W. (2008). Communication As Culture, Revised Edition: Essays on Media and Society. London, UNITED KINGDOM: Routledge. Retrieved from http://ebookcentral.proquest.com/lib/aul/detail.action?docID=356343

Carisimo, E., Galperin, H., & Alvarez-Hamelin, J. I. (2015). A new intrinsic way to measure IXP performance: an experience in Bolivia. ArXiv:1505.00837 [Cs]. Retrieved from http://arxiv.org/abs/1505.00837

304

Cerf, V. G., & Kahn, R. E. (1974). A Protocol for Packet Network Intercommunication. IEEE, (5), 13.

Chatzis, N., Smaragdakis, G., Feldmann, A., & Willinger, W. (2013). There is More to IXPs Than Meets the Eye. SIGCOMM Comput. Commun. Rev., 43(5), 19–28. https://doi.org/10.1145/2541468.2541473

Cheng, A.-S., & Fleischmann, K. R. (2010). Developing a meta-inventory of human values. Proceedings of the American Society for Information Science and Technology, 47(1), 1– 10. https://doi.org/10.1002/meet.14504701232

Cisco. (2016, September 12). BGP Best Path Selection Algorithm. Retrieved November 25, 2018, from https://www.cisco.com/c/en/us/support/docs/ip/border-gateway-protocol- bgp/13753-25.html

Clark, D. D. (1982, July). RFC 814 - Name, addresses, ports, and routes. Retrieved from https://tools.ietf.org/html/rfc814

Clark, D. D. (1988). The Design Philosophy of the DARPA Internet Protocols. Computer Communication Review, 10.

Clark, D. D. (2018). Designing an Internet (1st edition). Cambridge, MA: MIT Press.

Cockburn, C. (1983). Caught in the Wheels. Marxism Today, 16–21.

Cohen, J. E. (2012). Configuring the Networked Self: Law, Code, and the Play of Everyday Practice. New Haven Conn.: Yale University Press.

Columbia University. (2015). Proceedings of the Conference on Internet Governance and Cyber Security (p. 110). New York. N. Y.: Columbia University’s School of Public and International Affairs (SIPA) and Global Commission on Internet Governance (GCIG). Retrieved from https://sipa.columbia.edu/sites/default/files/Proceedings_ColumbiaSIPA_InternetGoveran ce_Cybersecurity_Conference2015.pdf

305

Dagmar Dehmer, V., & Haselberger, S. (2015). Die Chronologie des Spionageskandals. Der TagesSpiegel. Retrieved from https://www.tagesspiegel.de/politik/bnd-und-nsa-die- chronologie-des-spionageskandals/11722810.html

Daskal, J. C. (2018). Borders and Bits (SSRN Scholarly Paper No. ID 3096330). Rochester, NY: Social Science Research Network. Retrieved from https://papers.ssrn.com/abstract=3096330

Davidson, J., Hathaway, W., Postel, J., Mimno, N., Thomas, R., & Walden, D. (1977). The Arpanet Telnet Protocol: Its Purpose, Principles, Implementation, and Impact on Host Operating System Design. In Proceedings of the Fifth Symposium on Data Communications (pp. 4.10–4.18). New York, NY, USA: ACM. https://doi.org/10.1145/800103.803338

DC3. (2017, July). DC3 - Working Definitions and Principles - Declaration on Community Connectivity. Retrieved April 27, 2018, from https://comconnectivity.org/article/dc3- working-definitions-and-principles/

De Filippi, P., & Tréguer, F. (2016). Wireless Community Networks: Towards a Public Policy for the Network Commons? Hal Archives-Ouvertes.Fr, 1–17.

DE-CIX. (2016, September 16). Information on the lawsuit against the Federal Republic of Germany. Retrieved April 21, 2018, from https://www.de-cix.net/en/about-de-cix/media- center/press-releases/information-on-the-lawsuit-against-the-federal-republic-of-germany

DE-CIX. (n.d.-a). GlobePEER Remote - DE-CIX. Retrieved April 20, 2019, from https://www.de-cix.net/en/de-cix-service-world/globepeer-remote

DE-CIX. (n.d.-b). Traffic routing basics (BGP focus) - White paper. DE-CIX. Retrieved from https://www.de-cix.net/en/de-cix-service-world/globepeer/traffic-routing-basics

Deibert, R. J., & Crete-Nishihata, M. (2012). Global Governance and the Spread of Cyberspace Controls. Global Governance, 18(3), 339–361.

DeNardis, L., & Hackl, A. M. (2015). Internet governance by social media platforms. Telecommunications Policy, 39(9), 761–770. https://doi.org/10.1016/j.telpol.2015.04.003

306

DeNardis, L. (2009). Protocol Politics: The Globalization of Internet Governance (First Edition, 1st Printing edition). Cambridge, Mass: The MIT Press.

DeNardis, L. (2010). The Emerging Field of Internet Governance, 21.

DeNardis, L. (2012). Hidden Levers of Internet Control. Information, Communication & Society, 15(5), 720–738. https://doi.org/10.1080/1369118X.2012.659199

DeNardis, L. (2014). The global war for Internet governance. New Haven: Yale University Press.

DeNardis, L., & Hackl, A. M. (2016). Internet control points as LGBT rights mediation. Information, Communication & Society, 19(6), 753–770. https://doi.org/10.1080/1369118X.2016.1153123

DeNardis, L., & Musiani, F. (2016). Governance by infrastructure. In Francesca Musiani, Denardis, Laura, D. L. Cogburn, & Levinson, Nanette S. (Eds.), The turn to infrastructure in Internet governance (pp. 3–21). Palgrave Macmillan US.

Denning, P. J., & Martell, C. H. (2015). Great Principles of Computing. Cambridge, Massachusetts: The MIT Press.

Deutscher Bundestag (Ed.). (2015, March 26). Stenografisches Protokoll der 43 Sitzung - Endgültige Fassung* -. Deutscher Bundestag. Retrieved from http://dipbt.bundestag.de/dip21/btd/18/CD12850/D_I_Stenografische_Protokolle/Protoko ll%2043%20I.pdf

Dhamdhere, A., & Dovrolis, C. (2010). The Internet is Flat: Modeling the Transition from a Transit Hierarchy to a Peering Mesh. In Proceedings of the 6th International COnference (pp. 21:1–21:12). New York, NY, USA: ACM. https://doi.org/10.1145/1921168.1921196

Digital Equipment Corporation, Intel Corporation, & Xerox Corporation. (1982, November). The Ethernet - A Local Area Network, Data Link Layer and Physical Layer Specification - Version 2.0. Digital Equipmet, Intel Corp., Xerox Corp. Retrieved from http://decnet.ipv7.net/docs/dundas/aa-k759b-tk.pdf

307

D’Ignazio, A., & Giovannetti, E. (2009). Asymmetry and discrimination in Internet peering: evidence from the LINX. International Journal of Industrial Organization, 27(3), 441– 448. https://doi.org/10.1016/j.ijindorg.2008.11.004

Dimitropoulos, X., Krioukov, D., Fomenkov, M., Huffaker, B., Hyun, Y., claffy, kc, & Riley, G. (2007). AS Relationships: Inference and Validation. ACM SIGCOMM Computer Communication Review, 37(1), 29. https://doi.org/10.1145/1198255.1198259

Doneda, D., & Almeida, V. A. F. (2015). Privacy Governance in Cyberspace. IEEE Internet Computing, 19(3), 50–53. https://doi.org/10.1109/MIC.2015.66

Dordal, P. L. (2018). An Introduction to Computer Networks - Release 1.9.15. Chicago: Loyola University Chicago. Retrieved from http://intronetworks.cs.luc.edu/current/ComputerNetworks.pdf

Dourish, P. (2015). Protocols, Packets, and Proximity: The Materiality of Internet Routing. In L. Parks & N. Starosielski (Eds.), Signal Traffic: Critical Studies of Media Infrastructures (pp. 183–204). Urbana: University of Illinois Press. Retrieved from http://www.jstor.org/stable/10.5406/j.ctt155jmd9

Edmundson, A., Ensafi, R., Feamster, N., & Rexford, J. (2016). Characterizing and Avoiding Routing Detours Through Surveillance States. ArXiv:1605.07685 [Cs]. Retrieved from http://arxiv.org/abs/1605.07685

EFF. (n.d.). CALEA. Retrieved May 18, 2019, from https://www.eff.org/issues/calea

Ermert, M. (2013, July 2). PRISM scandal: internet exchange points as targets for surveillance - The H Security: News and Features. Retrieved September 17, 2018, from http://www.h- online.com/news/item/PRISM-scandal-internet-exchange-points-as-targets-for- surveillance-1909989.html

Euro-IX. (2015). euro-ix Internet Exchange Points: 2015 Report (p. 21). European Internet Exchange Association (Euro-IX). Retrieved from https://www.euro- ix.net/media/filer_public/11/2c/112c450a-7f3c-4947-9c0b-6104b3a38659/euro-ix-ixp- report-2015-final.pdf

308

Euro-IX. (n.d.). IXP Database. Retrieved April 2, 2019, from https://ixpdb.euro-ix.net/en/ix- f/ixp-database/

Evans, L. (2016, October 4). Pe•que•nos pro•ve•do•res estão em al•ta na internet. Estado de Minas. Retrieved from https://www.em.com.br/app/noticia/economia/2016/10/04/internas_economia,810644/pe- que-nos-pro-ve-do-res-em-al-ta.shtml

Fanou, R., Francois, P., Aben, E., Mwangi, M., Goburdhan, N., & Valera, F. (2017). Four years tracking unrevealed topological changes in the African interdomain. Computer Communications, 106, 117–135. https://doi.org/10.1016/j.comcom.2017.02.014

Fanou, R., Valera, F., Francois, P., & Dhamdhere, A. (2017). Reshaping the African Internet: From scattered islands to a connected continent. Computer Communications, 113, 25–42. https://doi.org/10.1016/j.comcom.2017.09.006

Fanou, Rodérick, Tyson, G., Francois, P., & Sathiaseelan, A. (2016). Pushing the Frontier: Exploring the African Web Ecosystem (pp. 435–445). ACM Press. https://doi.org/10.1145/2872427.2882997

Faratin, P., Clark, D. D., Bauer, S., Lehr, W., Gilmore, P. W., & Berger, A. (2008). The Growing Complexity of Internet Interconnection. Communications & Strategies, 72(4th Quarter), 51–71.

FCC. (2019, April 29). Communications Assistance for Law Enforcement Act. Retrieved May 18, 2019, from https://www.fcc.gov/public-safety-and-homeland-security/policy-and- licensing-division/general/communications-assistance

Fisher, T. (2018, March 23). What Are Hops & Hop Counts? Retrieved May 24, 2018, from https://www.lifewire.com/what-are-hops-hop-counts-2625905

Fleischmann, K. R. (2006). Boundary Objects with Agency: A Method for Studying the Design– Use Interface. The Information Society, 22(2), 77–87. https://doi.org/10.1080/01972240600567188

309

Florentino, A. (2018, August 28). Atributos BGP. NetFindersBrasil. Retrieved from https://www.youtube.com/channel/UCb6yTRhs-uwbeFnBf2-hlmw

Flores, M. C., & Glez, J. W. R. (2015, February). Distribución de la Población por Tamaño de Localidad y su Relación con el Medio Ambiente. Presented at the Seminario - taller “Información para la toma de decisiones: Población y Medio Ambiente,” Mexico City. Retrieved from https://www.inegi.org.mx/eventos/2015/Poblacion/doc/p- WalterRangel.pdf

Flores-Roux, E. M., Mariscal, J., & Aldama, F. A. (2009). Propuesta de licitación de la fibra oscura propiedad de la CFE: solución que genera escasez artificial, tanto presente como futura. DIRSI, TELECOM-CIDE, IDRC-CRDI, 1–45.

Foditsch, N. (2017). The Power of Airwaves: The Role of Spectrum Management in Media Development (p. 20). Washington, DC: Center for International Media Assistance National Endowment for Democracy. Retrieved from https://www.cima.ned.org/wp- content/uploads/2017/04/CIMA-Spectrum-Management_web.pdf

Foucault, M. (2008). Nascimento da biopolítica: curso dado no Collège de France (1978-1979). São Paulo: Martins Fontes.

Frey, K. (2000). Políticas Públicas: Um Debate Conceitual e Reflexões Referentes à Prática de da Análise de Políticas Públicas no Brasil. Planejamento e Políticas Públicas, 21(1), 211–259.

Friedman, B., Kahn, P. H., & Borning, A. (2006). Value Sensitive Design and Information Systems. In P. Zhang & D. F. Galletta (Eds.), Human-computer Interaction and Management Information Systems: Foundations (1 edition, pp. 1–27). London New York: Routledge.

Fuller, S. (2006). The Philosophy of Science and Technology Studies (1 edition). New York: Routledge.

310

Fuller, V., & Li, T. (2006, August). RFC 4632 - Classless Inter-domain Routing (CIDR): The Internet Address Assignment and Aggregation Plan. Internet Engineering Task Force (IEFT). Retrieved from https://tools.ietf.org/html/rfc4632

Galperín, H. (2016). Localizing Internet infrastructure: Cooperative peering in Latin America. Telematics and Informatics, 33(2), 631–640. https://doi.org/10.1016/j.tele.2015.08.010

Galperin, H., & Mariscal, J. (2016). Internet y pobreza: Evidencia y nuevas líneas de investigación para América Latina. CIDE.

Geist, A., Gjerding, S., & Moltke, H. (2014, October 20). German disclosure raises questions about Danish NSA-partnership. Retrieved May 13, 2019, from https://www.information.dk/udland/2014/10/german-disclosure-raises-questions-about- danish-nsa-partnership

Germany. (2009, August 14). Federal Data Protection Act (BDSG). Federal Ministry of Justice, Consumer Protection, and the Federal Office of Justice. Retrieved from https://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=6&ved=2ahUKE wj7k- 7fmsniAhWsILkGHVgsBJQQFjAFegQIABAC&url=https%3A%2F%2Fwww.legislatio nline.org%2Fdocuments%2Fid%2F6750&usg=AOvVaw2jOoctgXLYxLZOVghQkq0-

Germany. (2014, December 23). Basic Law for the Federal Republic of Germany. Federal Ministry of Justice, Consumer Protection, and the Federal Office of Justice. Retrieved from https://www.gesetze-im-internet.de/englisch_gg/englisch_gg.pdf

Germany. (2016, June 28). BND Act. Reporters without Borders. Retrieved from https://www.reporter-ohne- grenzen.de/fileadmin/Redaktion/user_upload/BNDGE_English_Translation_by_RSF.pdf

Germany. (2017, June 30). Federal Data Protection Act (BDSG). Federal Ministry of Justice, Consumer Protection, and the Federal Office of Justice. Retrieved from https://www.gesetze-im-internet.de/englisch_bdsg/englisch_bdsg.pdf

311

Gil, G., & Vimercati, L. (2007, November 13). Joint Declaration on Internet Rights. Retrieved from https://www.civile.it/mobile/visual.php?num=50532

Gill, P., Arlitt, M., Li, Z., & Mahanti, A. (2008). The Flattening Internet Topology: Natural Evolution, Unsightly Barnacles or Contrived Collapse? In Proceedings of the 9th International Conference on Passive and Active Network Measurement (pp. 1–10). Berlin, Heidelberg: Springer-Verlag. Retrieved from p. 8

Gillespie, T. (2010). The politics of ‘platforms.’ New Media & Society, 12(3), 347–364. https://doi.org/10.1177/1461444809342738

Gilliom, J. (2001). Overseers of the Poor: Surveillance, Resistance, and the Limits of Privacy. University of Chicago Press.

Giotsas, V., Zhou, S., Luckie, M., & claffy, kc. (2013). Inferring multilateral peering (pp. 247– 258). Presented at the ACM CoNEXT, ACM Press. https://doi.org/10.1145/2535372.2535390

Glaser, B. G., & Strauss, A. L. (2006). The discovery of grounded theory: strategies for qualitative research (1st (reprinted)). New Brunswick: Aldine.

Glaser, H. R., & Canabarro, D. R. (2016). Before and after the WGIG: Twenty years of multistakeholder Internet governance in Brazil. In W. J. Drake (Ed.), The working group on Internet governance: 10th anniversary reflection (1st Edition, p. 258). Uruguay: APC. Retrieved from https://www.apc.org/sites/default/files/IG_10_Final_0.pdf

Gourdin, E., Maille, P., Simon, G., & Tuffin, B. (2017). The Economics of CDNs and Their Impact on Service Fairness. IEEE Transactions on Network and Service Management, 14(1), 22–33. https://doi.org/10.1109/TNSM.2017.2649045

Greenwald, G. (2014). No Place to Hide: Edward Snowden, the NSA, and the U.S. Surveillance State (1st edition). New York: Metropolitan Books.

Grosfoguel, R. (2011). Decolonizing Post-Colonial Studies and Paradigms of Political-Economy: Transmodernity, Decolonial Thinking, and Global Coloniality. TRANSMODERNITY:

312

Journal of Peripheral Cultural Production of the Luso-Hispanic World, 1(1). Retrieved from https://escholarship.org/uc/item/21k6t3fq#author

Grossmann, L. O. (2018a, July 4). Pequenos provedores concentram dois terços dos novos acessos internet no Brasil. Convergência Digital. Retrieved from http://www.convergenciadigital.com.br/cgi/cgilua.exe/sys/start.htm?UserActiveTemplate =site&infoid=48384&sid=4

Grossmann, L. O. (2018b, November 1). Anatel unifica conceito de pequeno provedor em todos os regulamentos. Convergência Digital. Retrieved from http://www.convergenciadigital.com.br/cgi/cgilua.exe/sys/start.htm?UserActiveTemplate =site&infoid=49375&sid=4

Gupta, A., Vanbever, L., Shahbaz, M., Donovan, S. P., Schlinker, B., Feamster, N., … Katz- Bassett, E. (2014). SDX: A Software Defined Internet Exchange. In Proceedings of the 2014 ACM Conference on SIGCOMM (pp. 551–562). New York, NY, USA: ACM. https://doi.org/10.1145/2619239.2626300

Haggerty, K. D., & Ericson, R. V. (2003). The surveillant assemblage. The British Journal of Sociology, 51(4), 605–622. https://doi.org/10.1080/00071310020015280

Hall, R. (2015). Terror and the Female Grotesque: Introducing Full-Body Scanners to U.S. Airports. In R. E. Dubrofsky & S. A. Magnet (Eds.), Feminist Surveillance Studies (Edition Unstated edition, pp. 127–149). Durham: Duke University Press Books.

Haraway, D. (1988). Situated Knowledges: The Science Question in Feminism and the Privilege of Partial Perspective. Feminist Studies, 14(3), 575–599. https://doi.org/10.2307/3178066

Harding, S. (2016). Latin American Decolonial Social Studies of Scientific Knowledge: Alliances and Tensions. Science, Technology, & Human Values, 41(6), 1063–1087. https://doi.org/10.1177/0162243916656465

Harding, S. (2017). Latin American Decolonial Studies: Feminist Issues, 14.

313

Hawkinson, J., & Bates, T. (1996, March). RFC 1930 - Guidelines for creation, selection, and registration of an Autonomous System (AS). Retrieved from https://tools.ietf.org/html/rfc1930#page-2

Hayles, N. K. (1999). How We Became Posthuman: Virtual Bodies in Cybernetics, Literature, and Informatics. University of Chicago Press.

Hecht, G. (Ed.). (2011). Entangled Geographies: Empire and Technopolitics in the Global Cold War (1st Edition). Cambridge, MA: The MIT Press.

Hepp, A. (2015). Transcultural Communication (1st Edition). Chichester, West Sussex; Malden, MA: John Wiley & Sons.

Hill, J. (2012). Internet Fragmentation: Highlighting the Major Technical, Governance and Diplomatic Challenges for U.S. Policy Makers (SSRN Scholarly Paper No. ID 2439486). Rochester, NY: Social Science Research Network. Retrieved from https://papers.ssrn.com/abstract=2439486

Huerta, E. (2016). Manual de Telefonía Celular Comunitaria: Conectando al Seguiente Billión. Redes por la Diversidad, Equidad y Sustentabilidad A.C. Retrieved from http://media.wix.com/ugd/68af39_dbb75b7324284194b12acb56ec8210db.pdf

Huerta, E. (2018). La Compartencia: Bases para la Formación en Comunicación Indígena (Ph.D. in Rural Development). Universidad Autónoma Metropolitana, Mexico City.

Hughes, M., Pels, M., & Michl, H. (2013, November). Internet Exchange Point “Wishlist” - Version 4.0.4. euro-ix.

IEEE. (2016, March 4). IEEE Standard for Ethernet. Retrieved from https://ieeexplore.ieee.org/document/7428776/definitions#definitions

IFT. (2016). P/IFT/010716/349: RESOLUCIÓN MEDIANTE LA CUAL EL PLENO DEL INSTITUTO FEDERAL DE TELECOMUNICACIONES OTORGA UN TÍTULO DE CONCESIÓN PARA USAR Y APROVECHAR BANDAS DE FRECUENCIAS DEL ESPECTRO RADIOELÉCTRICO, ASÍ COMO UN TÍTULO DE CONCESIÓN ÚNICA, AMBOS PARA USO SOCIAL INDÍGENA, EN FAVOR DE

314

TELECOMUNICACIONES INDÍGENAS COMUNITARIAS, A.C. Instituto Federal de Telecomunicaciones. Retrieved from http://apps.ift.org.mx/publicdata/P_IFT_010716_349_FR_P19y20.pdf

IFT. (2017). Tercer Informe Trimestral Estadístico 2016 (p. 91). Mexico City: Instituto Federal de Telecomunicaciones.

IFT - Consejo Consultivo. (2015). Recomendación del Consejo Consultivo Del Instituto Federal de Telecomunicaciones sobre las Manifestaciones de Interés para el Desarollo de Red Compartida Mayorista. IFT. Retrieved from http://consejoconsultivo.ift.org.mx/docs/recomendaciones/Recomendaci%C3%B3n,%20 Manifestaciones%20de%20intereses%20(MDI).pdf

INEGI. (2016). Estadísticas a Propósito del Día Internacional de los Pueblos Indígenas (9 de Agosto) - Datos Nacionales (p. 14). Aguas Calientes: INEGI - Instituto Nacional de Estadística y Geografía.

Intven, H., Oliver, J., & Sepúlveda, E. (2000). Telecommunications Regulation Handbook. (H. Intven & M. Tétrault, Eds.). Washington, DC: infoDev, The World Bank. Retrieved from http://hdl.handle.net/10986/15249

ITU. (n.d.). ITU EMF Guide. Retrieved February 14, 2019, from http://emfguide.itu.int/en/emfguide_m.html izzi. (2014, February 7). Contrato de prestación de servicios. Retrieved from https://www.izzi.mx/registro/ajax/legal

Jackson, S. J., Edwards, P. N., Bowker, G. C., & Knobel, C. P. (2007). Understanding infrastructure: History, heuristics and cyberinfrastructure policy. First Monday, 12(6). https://doi.org/10.5210/fm.v12i6.1904

JafariNaimi, N., Nathan, L., & Hargraves, I. (2015). Values as Hypotheses: Design, Inquiry, and the Service of Values. Design Issues, 31(4), 91–104. https://doi.org/10.1162/DESI_a_00354

315

Jain, R. (2015). A Model for Internet Governance and Implications for India (IIMA Working Papers No. WP2015- 03–23). Indian Institute of Management Ahmedabad, Research and Publication Department. Retrieved from https://ideas.repec.org/p/iim/iimawp/13332.html

Jasanoff, S. (Ed.). (2004). States of knowledge: the co-production of science and social order. London ; New York: Routledge.

Joerges, B. (1999). Do Politics Have Artefacts? Social Studies of Science, 29(3), 411–431. https://doi.org/10.1177/030631299029003004

Karasti, H. (2014). Infrastructuring in Participatory Design. In Proceedings of the 13th Participatory Design Conference: Research Papers - Volume 1 (pp. 141–150). New York, NY, USA: ACM. https://doi.org/10.1145/2661435.2661450

Karlin, J., Forrest, S., & Rexford, J. (2009). Nation-State Routing: Censorship, Wiretapping, and BGP. ArXiv:0903.3218 [Cs]. Retrieved from http://arxiv.org/abs/0903.3218

Katz, M. L., & Shapiro, C. (1994). Systems Competition and Network Effects. The Journal of Economic Perspectives, 8(2), 93–115.

Key4biz. (2005, October 12). Tunisi Mon Amour. Cortiana: ´Per una Dichiarazione dei Diritti della Rete´. Retrieved May 30, 2019, from https://www.key4biz.it/News-2005-10-12- Policy-Tunisi-Mon-Amour-Cortiana-Per-una-Dichiarazione-dei-Diritti-della-Rete- 154096/60207/

Klein, M. (2009). Wiring Up The Big Brother Machine...And Fighting It. Charleston, S.C: BookSurge Publishing.

Klöti, R., Ager, B., Kotronis, V., Nomikos, G., & Dimitropoulos, X. (2016). A Comparative Look into Public IXP Datasets. ArXiv:1611.02624 [Cs]. Retrieved from http://arxiv.org/abs/1611.02624

Kommers, D. P., & Miller, R. A. (2012). The Constitutional Jurisprudence of the Federal Republic of Germany: Third Edition, Revised and Expanded (3rd Edition). Durham: Duke University Press.

316

Kotronis, V., Klöti, R., Rost, M., Georgopoulos, P., Ager, B., Schmid, S., & Dimitropoulos, X. (2016). Stitching Inter-Domain Paths over IXPs. In Proceedings of the Symposium on SDN Research (pp. 17:1–17:12). New York, NY, USA: ACM. https://doi.org/10.1145/2890955.2890960

Kuerbis, B., & Mueller, M. (2017). Internet routing registries, data governance, and security. Journal of Cyber Policy, 2(1), 64–81. https://doi.org/10.1080/23738871.2017.1295092

Labovitz, C., Iekel-Johnson, S., McPherson, D., Oberheide, J., & Jahanian, F. (2010). Internet Inter-domain Traffic. In Proceedings of the ACM SIGCOMM 2010 Conference (pp. 75– 86). New York, NY, USA: ACM. https://doi.org/10.1145/1851182.1851194

Larkin, B. (2013). The Politics and Poetics of Infrastructure. Annual Review of Anthropology, 42(1), 327–343. https://doi.org/10.1146/annurev-anthro-092412-155522

Latour, B. (1991). Technology Is Society Made Durable. The Sociological Review Monograph, (38), 103–131.

Latour, B. (1999). Pandora’s Hope: Essays on the Reality of Science Studies (1 edition). Cambridge, Mass: Harvard University Press.

Latour, B. (2002). Morality and Technology. Theory, Culture & Society, 19(5–6), 247–260. https://doi.org/10.1177/026327602761899246

Latour, B. (2005). Reassembling the Social: An Introduction to Actor-Network-Theory (1st edition). Oxford; New York: Oxford University Press.

Latour, B. (2007). A Textbook Case Revisited – Knowledge as a Mode of Existence. In E. J. Hackett, O. Amsterdamska, M. E. Lynch, & J. Wajcman (Eds.), The Handbook of Science and Technology Studies (3rd edition, pp. 83–112). Cambridge, Mass: The MIT Press.

Latour, B. (2008). Where Are the Missing Masses? The Sociology ofa Few Mundane Artifacts. In D. J. Johnson & J. M. Wetmore (Eds.), Technology and Society, Building Our Sociotechnical Future (1st Edition, pp. 151–180). Cambridge, MA: MIT Press.

317

Latour, B. (2013). An Inquiry into Modes of Existence: An Anthropology of the Moderns. (C. Porter, Trans.) (1st Edition). Cambridge, Massachusetts: Harvard University Press.

Law, J. (2015, June 24). STS as Method. Retrieved from http://heterogeneities.net/publications/Law2015STSAsMethod.pdf

Le Dantec, C. A., Poole, E. S., & Wyche, S. P. (2009). Values as lived experience: evolving value sensitive design in support of value discovery. In Proceedings of the 27th international conference on Human factors in computing systems - CHI 09 (p. 1141). Boston, MA, USA: ACM Press. https://doi.org/10.1145/1518701.1518875

Lee, H., Lee, D., & Yi, Y. (2014). On the economic impact of Telco CDNs and their alliance on the CDN market. In 2014 IEEE International Conference on Communications (ICC) (pp. 2950–2955). https://doi.org/10.1109/ICC.2014.6883773

Lepido, D. (2019, April 30). Vodafone Found Hidden Backdoors in Huawei Equipment. Bloomberg. Retrieved from https://www.bloomberg.com/news/articles/2019-04- 30/vodafone-found-hidden-backdoors-in-huawei-equipment

Lessig, L. (1999). Open Code and Open Societies: Values of Internet Governance. Chicago-Kent Law Review, 74(3), 17.

Lessig, L. (2006). Code: Version 2.0 (2nd Revised ed. edition). New York: Basic Books.

Levinson, N. S., & Cogburn, D. L. (2016). The Next “Turn” in Internet Infrastructure Governance. In Francesca Musiani, Denardis, Laura, D. L. Cogburn, & Levinson, Nanette S. (Eds.), The Turn to Infrastructure in Internet Governance (1st Edition, pp. 219–223). New York: Palgrave Macmillan. Retrieved from //www.palgrave.com/br/book/9781137533265

Lievrouw, L. A. (2014). Materiality and media in communication and technology studies: An unfinished project. In T. Gillespie (Ed.), Media technologies: Essays on communication, materiality, and society (pp. 21–51).

Lievrouw, L. A., & Livingstone, S. (2010a). Introduction to the First Edition (2002) the Social Shaping and Consequences of ICTs. In Handbook of New Media: Social Shaping and

318

Social Consequences of ICTs, Updated Student Edition (pp. 15–32). London: SAGE Publications Ltd. https://doi.org/10.4135/9781446211304

Lievrouw, L. A., & Livingstone, S. (2010b). Introduction to the Updated Student Edition. In Handbook of New Media: Social Shaping and Social Consequences of ICTs, Updated Student Edition (pp. 1–14). 1 Oliver’s Yard, 55 City Road, London EC1Y 1SP United Kingdom: SAGE Publications Ltd. https://doi.org/10.4135/9781446211304.n1

Lingel, J., & Sinnreich, A. (2016). Incoded counter-conduct: What the incarcerated can teach us about resisting mass surveillance. First Monday, 21(5). Retrieved from http://journals.uic.edu/ojs/index.php/fm/article/view/6172

Lodhi, A., Larson, N., Dhamdhere, A., Dovrolis, C., & claffy, kc. (2014). Using peeringDB to understand the peering ecosystem. ACM SIGCOMM Computer Communication Review, 44(2), 20–27. https://doi.org/10.1145/2602204.2602208

Lopez, A. J. (2007). Introduction: The (Post) Global South. The Global South, 1(1), 1–11.

Lugones, M. (2010). Toward a Decolonial Feminism. Hypatia, 25(4), 742–759. https://doi.org/10.1111/j.1527-2001.2010.01137.x

Lumumba-Kasongo, T. (2015). Rethinking the Bandung conference in an Era of ‘unipolar liberal globalization’ and movements toward a ‘multipolar politics.’ Bandung: Journal of the Global South, 2(1). https://doi.org/10.1186/s40728-014-0012-4

MacKenzie, D., & Wajcman, J. (1999). Introductory essay: the social shaping of technology. In D. MacKenzie & J. Wajcman (Eds.), The social shaping of technology (2nd ed., pp. 3– 27). Buckingham, UK: Open University Press.

Magnet, S. (2011). When biometrics fail: gender, race, and the technology of identity. Durham: Duke University Press.

Mahler, A. G. (2017, October 25). Global South - Literary and Critical Theory - Oxford Bibliographies - obo. Oxford Bibliographies. Retrieved from http://www.oxfordbibliographies.com/view/document/obo-9780190221911/obo- 9780190221911-0055.xml

319

Maigron, P. (2019, April 8). Regional Internet Registries Statistics. Retrieved April 14, 2019, from https://www-public.imtbs- tsp.eu/~maigron/RIR_Stats/RIR_Delegations/World/ASN-ByNb.html

Marthews, A., & Tucker, C. E. (2017). Government Surveillance and Internet Search Behavior (SSRN Scholarly Paper No. ID 2412564). Rochester, NY: Social Science Research Network. Retrieved from https://papers.ssrn.com/abstract=2412564

Martínez Luna, J. (2010). Eso que llaman comunalidad. Mexico: CONACULTA, Secretaría de Cultura de Oaxaca, Campo.

Mascolo, V. G., Leyendecker, H., & Goetz, J. (2014, October 4). Codewort Eikonal - der Albtraum der Bundesregierung. sueddeutsche.de. Retrieved from https://www.sueddeutsche.de/politik/geheimdienste-codewort-eikonal-der-albtraum-der- bundesregierung-1.2157432

Massanari, A. (2017). #Gamergate and The Fappening: How Reddit’s algorithm, governance, and culture support toxic technocultures. New Media & Society, 19(3), 329–346. https://doi.org/10.1177/1461444815608807

Mathew, A. J. (2014). Where in the World is the Internet? Locating Political Power in Internet Infrastructure (Dissertation). University of California – Berkeley. Retrieved from https://www.ischool.berkeley.edu/sites/default/files/ashwin-dissertation.pdf

Mathew, A. J. (2016). The myth of the decentralised internet. Internet Policy Review. Retrieved from https://policyreview.info/articles/analysis/myth-decentralised-internet

Maxwell, J. A. (2013). Qualitative research design: an interactive approach (3rd ed.). Thousand Oaks, Calif: Sage Publications.

Mcdonald, B. P. (2004). The First Amendment and the Free Flow of Information: Towards A Realistic Right to Gather Information in the Information Age. Ohio State Law Journal, 65(2), 249–356.

320

McKeown, N., Anderson, T., Balakrishnan, H., Parulkar, G., Peterson, L., Rexford, J., … Turner, J. (2008). OpenFlow: enabling innovation in campus networks. ACM SIGCOMM Computer Communication Review, 38(2), 69. https://doi.org/10.1145/1355734.1355746

Meier-Hahn, U. (2014a). Internet exchanges as organisers in the interconnection market (p. 21). Presented at the Score International Conference on Organizing Markets, Stockholm, Sweden.

Meier-Hahn, U. (2014b). Internet Interconnection: How the Economics of Convention Can Inform the Discourse on Internet Governance (p. 22). Presented at the GigaNet: Global Internet Governance Academic Network, Annual Symposium 2014, Rochester, NY: Social Science Research Network. Retrieved from https://papers.ssrn.com/abstract=2809867

Meister, A. (2015, March 31). How the German Foreign Intelligence Agency BND tapped the Internet Exchange Point DE-CIX in Frankfurt, since 2009. Retrieved April 21, 2018, from https://netzpolitik.org/2015/how-the-german-foreign-intelligence-agency-bnd- tapped-the-internet-exchange-point-de-cix-in-frankfurt-since-2009/

Mendonza Bautista, D. J. (2017). Servicio de Carrera en el Régimen de Usos y Costumbres de los Municipios de la Sierra Juárez de Oaxaca (Undergraduate Thesis of Political Science and Public Administration). Universidad Nacional Autónoma de México, Mexico City.

MERIT. (1994, July). From the ARPANET to the World’s Largest and Fastest Network for Research and Education. Link Letter - The Merit/ NSF Project. Retrieved from https://hdl.handle.net/2027/mdp.39015035356347

Metcalfe, R. M., & Boggs, D. R. (1976). Ethernet: distributed packet switching for local computer networks. Communications of the ACM, 19(7), 395–404. https://doi.org/10.1145/360248.360253

Metz, C. (2001). Interconnecting ISP Networks. IEEE Internet Computing, 5(2), 74–80. https://doi.org/10.1109/4236.914650

321

Mexico. (1917, 2018). Mexico’s Constitution of 1917 with Amendments through 2015. (M. F. G. Aban, Trans.). constituteproject.org. Retrieved from https://www.constituteproject.org/constitution/Mexico_2015.pdf?lang=en

Mexico. (2014, July 14). Ley Federal de Telecomunicaciones y Radiodifusión.

Mexico. (2017, July 24). DOF - Diario Oficial de la Federación. Retrieved from http://www.dof.gob.mx/nota_detalle.php?codigo=5491665&fecha=24/07/2017

Mexico. (2018). México - Presidencia de la República - Biografías de los gobernantes de México. Retrieved April 30, 2018, from http://fox.presidencia.gob.mx/mexico/gobernantes/?c=bio&bio=50

Mignolo, W. D. (2002). The Geopolitics of Knowledge and the Colonial Difference. South Atlantic Quarterly, 101(1), 57–96. https://doi.org/10.1215/00382876-101-1-57

Mitchell, B. (2019, May 13). Port Numbers Used for TCP/IP Computer Networks. Retrieved May 21, 2019, from https://www.lifewire.com/port-numbers-on-computer-networks- 817939

Moore, M., & Tambini, D. (Eds.). (2018). Digital Dominance: The Power of Google, Amazon, Facebook, and Apple (1st Edition). New York: Oxford University Press.

Motta, L. (n.d.). Ether. In Eric Weisstein’s World of Physics. Retrieved from http://scienceworld.wolfram.com/physics/Ether.html

Mueller, M. L. (2010). Networks and States: The Global Politics of Internet Governance. Cambridge, Mass: The MIT Press.

Muñoz, A. A. (2005). The Emergence and Development of the Politics of Recognition of Cultural Diversity and Indigenous Peoples’ Rights in Mexico: Chiapas and Oaxaca in Comparative Perspective. Journal of Latin American Studies, 37(3), 585–610.

Murphy, S. (2006, January). RFC 4272 - BGP Security Vulnerabilities Analysis. Retrieved from https://tools.ietf.org/html/rfc4272

Musiani, F. (2013). Network architecture as internet governance. Internet Policy Review, 2(4), 1– 9. 322

Musiani, F., Cogburn, D. L., DeNardis, L., & Levinson, N. S. (Eds.). (2016). The Turn to Infrastructure in Internet Governance. New York: Palgrave Macmillan US.

Nagy, P., & Neff, G. (2015). Imagined Affordance: Reconstructing a Keyword for Communication Theory. Social Media + Society, 1(2), 205630511560338. https://doi.org/10.1177/2056305115603385

Neal, M. (2013, July 22). Here’s What Wi-Fi Would Look Like If We Could See It. Vice. Retrieved from https://www.vice.com/en_us/article/9an9m7/heres-what-wi-fi-would- look-like-if-we-could-see-it

NETmundial. (2014a, April). Links to Internet Governance Principles. Retrieved May 30, 2019, from http://netmundial2014.org

NETmundial. (2014b, April). NETmundial: the beginning of a process. | Net Mundial. Retrieved May 30, 2019, from http://netmundial.br/about/

NETmundial. (2014c, April 24). NETmundial Multistakeholder Statement. NETmundial. Retrieved from http://netmundial.br/wp-content/uploads/2014/04/NETmundial- Multistakeholder-Document.pdf

NIC.br. (2014). Problemas e implicações do uso do CGNAT (NAT no provedor) na Internet. São Paulo. Retrieved from https://www.youtube.com/watch?v=uxu0ivaDAnc

NIC.br. (n.d.). Localidades Atuais do IX.br. Retrieved April 7, 2019, from http://ix.br/localidades/atuais

NIC.br, & IX.br. (2017, June 30). Política de Requisitos Técnicos IX.br - V1.0. Retrieved from http://ix.br/doc/PRT_IX.br_V1.0_30_06_2017.pdf

Nissenbaum, H. (2009). Privacy in Context: Technology, Policy, and the Integrity of Social Life (1st Edition). Stanford, California: Stanford Law Books.

NITRD. (1991, April 30). Terms of Reference for the US Federal Engineering Planning Group (FEPG) Draft. The Networking and Information Technology Research and Development Program. Retrieved from https://www.nitrd.gov/fnc/FEPG.aspx

323

NITRD. (1995, December 11). Proposal for the Continued Interconnection of the Federal Agency Networks. NITRD. Retrieved from https://www.nitrd.gov/fnc/FEPG_proposal.aspx

NITRD. (1997). IV. CIC R&D Program Organization. The Networking and Information Technology Research and Development Program. Retrieved from https://www.nitrd.gov/PUBS/bluebooks/1997/organization.html

Nocetti, J. (2015). Contest and conquest: Russia and global internet governance. International Affairs, 91(1), 111–130. https://doi.org/10.1111/1468-2346.12189

Nomikos, G., & Dimitropoulos, X. (2016). traIXroute: Detecting IXPs in traceroute paths. ArXiv:1611.03895 [Cs]. Retrieved from http://arxiv.org/abs/1611.03895

Norton, W. B. (1999, May 21). Interconnection Strategies for ISPs - Document v2.1. Equinix. Retrieved from https://drpeering.net/white-papers/_pdfs/Interconnection-Strategies-for- ISPs.pdf

NSRC. (2016, January 18). BGP Best Current Practices - ISP Workshop. Network Startup Resource Center (NSRC).

Nye Jr., J. S. (2014). The Regime Complex For Managing Global Cyber Activities. Global Commission on Internet Governance, 5–15.

OECD. (2012). OECD Review of Telecommunication Policy and Regulation in Mexico. OECD Publishing. https://doi.org/10.1787/9789264060111-en

OECD, & IDB. (2016). Broadband policies for Latin America and the Caribbean: a digital economy toolkit. Paris: OECD Publishing.

O’Flaherty, C. (2018). Construcción de infraestructura comunitaria: tecnologías y modelos disruptivos. In L. Belli & O. Cavalli (Eds.), Gobernanza y regulaciones de Internet en América Latina: Análisis sobre infraestructura, privacidad, ciberseguridad y evoluciones tecnológicas en honor de los diez años de la South School on Internet Governance (1st Edition, pp. 235–244). Rio de Janeiro, Brazil: Escola de Direito do Rio de Janeiro da Fundação Getulio Vargas.

324

Oppermann, D. (2018). From Bandung to the DNS. In D. Oppermann (Ed.), Internet Governance in the Global South (1st Edition, pp. 12–41). São Paulo: Núcleo de Pesquisa em Relações Internacionais (NUPRI), Universidade de São Paulo (USP). Retrieved from https://www.academia.edu/38174294/Internet_Governance_in_the_Global_South

Oracle. (n.d.-a). Protocol Layers and the OSI Model (System Administration Guide: IP Services). Retrieved November 8, 2018, from https://docs.oracle.com/cd/E19683-01/806- 4075/ipov-7/index.html

Oracle. (n.d.-b). TCP/IP Protocol Architecture Model (System Administration Guide: IP Services). Retrieved November 8, 2018, from https://docs.oracle.com/cd/E19683-01/806- 4075/ipov-10/index.html

Oxman, J. (1999, July). The FCC and the Unregulation of the Internet - OPP Working Paper No. 31. Federal Communications Commission. Retrieved from https://transition.fcc.gov/Bureaus/OPP/working_papers/oppwp31.pdf

Papacharissi, Z. (2010). Privacy as a luxury commodity. First Monday, 15(8). Retrieved from https://journals.uic.edu/ojs/index.php/fm/article/view/3075

Papier, H.-J. (2016). Beschränkungen der Telekommunikationsfreiheit durch den BND an Datenaustauschpunkten. Neue Zeitschrift für Verwaltungsrecht – Extra, 35(15), 1–15.

Parra, D. (2015). Alternative Media in Latin American Grassroots Integration: Building Networks and New Agendas. International Journal of Communication, 9, 3680–3701.

Perez, D. A. L., Brito, S. H. B., Fontes, R. dos R., & Rothenberg, C. E. (2016). Delivering Application-Layer Traffic Optimization Services based on Public Routing Data at Internet eXchange Points - Semantic Scholar (pp. 1–14). Presented at the 34th Brazilian Symposium on Computer Networks and Distributed Systems (SBRC), Salvador, Brazil. Retrieved from /paper/Delivering-Application-Layer-Traffic-Optimization-Perez- Brito/f6613ca51df6247ade68a272d7f55a5f964b0162

Postel, J. (1979, August). IEN 116 - Internet Name Server. Retrieved from https://www.rfc- editor.org/ien/ien116.txt

325

Postel, J. (1980, August 28). RFC 768 - User Datagram Protocol. Retrieved from https://tools.ietf.org/html/rfc768

Postel, J. (1981a, September). RFC 791 - Internet Protocol. Internet Engineering Task Force (IEFT). Retrieved from https://tools.ietf.org/html/rfc791

Postel, J. (1981b, September). RFC 793 - Transmission Control Protocol. Internet Engineering Task Force (IEFT). Retrieved from https://tools.ietf.org/html/rfc793

Poster, M. (1990). The Mode of Information: Poststructuralism and Social Context. University of Chicago Press.

Priest, D., & Arkin, W. M. (2011). Top Secret America: The Rise of the New American Security State (1 edition). New York: Little, Brown and Company.

Prior, L. (2008). Repositioning Documents in Social Research. Sociology, 42(5), 821–836.

Putta Venkata, R., & Ruan, L. (2016). A Region-Centric Analysis of the Internet Peering Ecosystem. Computer Science Technical Reports, 1–8.

Quijano, A. (2007). Coloniality and Modernity/Rationality. Cultural Studies, 21(2–3), 168–178. https://doi.org/10.1080/09502380601164353

Rajak, S. (2014). No Bargaining Chips, No Spheres of Interest: The Yugoslav Origins of Cold War Non-Alignment. Journal of Cold War Studies, 16(1), 146–179. https://doi.org/10.1162/JCWS_a_00434

Raymond, M., & DeNardis, L. (2015). Multistakeholderism: anatomy of an inchoate global institution. International Theory, 7(3), 572–616. https://doi.org/10.1017/S1752971915000081

Rekhter, Y., Li, T., & Hares, S. (2006, January). RFC 4271 - A Border Gateway Protocol 4 (BGP-4). The Internet Society. Retrieved from https://tools.ietf.org/html/rfc4271

Richter, P., Smaragdakis, G., Feldmann, A., Chatzis, N., Boettger, J., & Willinger, W. (2014). Peering at Peerings: On the Role of IXP Route Servers. In Proceedings of the 2014 Conference on Internet Measurement Conference (pp. 31–44). New York, NY, USA: ACM. https://doi.org/10.1145/2663716.2663757 326

RIPE NCC. (2008, March 17). YouTube Hijacking: A RIPE NCC RIS case study. Retrieved November 13, 2018, from https://www.ripe.net/publications/news/industry- developments/youtube-hijacking-a-ripe-ncc-ris-case-study

Rivera, A. (2014, May 2). Forman en KIO el primer IXP de México. Retrieved August 1, 2017, from http://boletin.com.mx/datacenter-almacenamiento-y-respaldo-boletin/item/755- forman-en-kio-el-primer-ixp-de-mexico/755-forman-en-kio-el-primer-ixp-de-mexico

Roberts, D. E. (1991). Punishing Drug Addicts Who Have Babies: Women of Color, Equality, and the Right of Privacy. Harvard Law Review, 104(7), 1419. https://doi.org/10.2307/1341597

Roberts, H., Larochelle, D., Faris, R., & Palfrey, J. (2011). Mapping Local Internet Control. Berkman Center for Internet & Society at Harvard University, 14.

Rosa, F., & Vicentin, D. (2016). Governança da Internet e suas Implicações para as Políticas Públicas. Critical Reviews on Latin American Research - CROLAR, 5(1). Retrieved from https://www.crolar.org/index.php/crolar/article/view/247

Rossini, C., Cruz, F. B., & Doneda, D. (2015). The Strengths and Weaknesses of the Brazilian Internet Bill of Rights: Examining a Human Rights Framework for the Internet. Global Commission on Internet Governance, 36.

Rouvroy, A., & Berns, T. (2018). Governamentalidade algorítmica e perspectivas de emancipação: o díspar como condição de individuação pela relação? In F. Bruno, B. Cardoso, M. Kanashiro, L. Guilhon, & L. Melgaço (Eds.), Tecnopolíticas da Vigilância. Perspectivas da Margem (1st ed., pp. 107–139). São Paulo, SP: Boitempo.

Ruano-Borbalan, J.-C. (2017). Technology, Science And Society: Norms, Cultures, and Institutions Matter. Journal of Innovation Economics & Management, (22), 3–8. https://doi.org/10.3917/jie.022.0003

Rutherford, J., Gillespie, A., & Richardson, R. (2004). The territoriality of Pan-European telecommunications backbone networks. Journal of Urban Technology, 11(3), 1–34. https://doi.org/10.1080/10630730500064166

327

Sandvine. (2018). The Global Internet Phenomena Report. Sandvine. Retrieved from https://www.sandvine.com/hubfs/downloads/phenomena/2018-phenomena-report.pdf

Santoro, M., & Borges, B. (2017). Brazilian Foreign Policy Towards Internet Governance. Revista Brasileira de Política Internacional, 60(1). https://doi.org/10.1590/0034- 7329201600111

Santos, B. de S. (1995). Toward a New Common Sense: Law, Science and Politics in the Paradigmatic Transition (1 edition). New York: Routledge.

Santos, B. de S. (2006). Globalizations. Theory, Culture & Society, 23(2–3), 393–399. https://doi.org/10.1177/026327640602300268

Santos, B. de S. (2012). Public Sphere and Epistemologies of the South. Africa Development / Afrique et Développement, 37(1), 43–67.

Santos, B. de S., & Meneses, M. P. (Eds.). (2009). Introdução. In Epistemologias do Sul (1st ed., pp. 9–19). Coimbra: Edições Almedina.

Sargsyan, T. (2016). The privacy role of information intermediaries through self-regulation. Internet Policy Review. Retrieved from https://policyreview.info/articles/analysis/privacy-role-information-intermediaries- through-self-regulation

Sassen, S. (2003). Globalization or denationalization? Review of International Political Economy, 10(1), 1–22. https://doi.org/10.1080/0969229032000048853

Schindler, J. (n.d.). Bad Aibling Station: A Legacy of Excellence. Retrieved from https://www.webcitation.org/5k8f6XouC

Schley, S. (2017, August 25). Exclusive Interview - Bob Metcalfe The Father of Ethernet. Retrieved November 21, 2018, from https://broadbandlibrary.com/bob-metcalfe-lays- down-the-law/

SEDESOL. (2005a). Catálogo Localidades - Información de localidad - Santa María Yaviche. Secretaria de Desarrollo Social. Retrieved from http://www.microrregiones.gob.mx/catloc/contenido.aspx?refnac=205410002 328

SEDESOL. (2005b). Catálogo Localidades - Resumen municipal - de Guelatao de Juárez. Secretaria de Desarrollo Social. Retrieved from http://www.microrregiones.gob.mx/catloc/LocdeMun.aspx?tipo=clave&campo=loc&ent =20&mun=035

SEDESOL. (2005c). Catálogo Localidades - Resumen municipal - Municipio de Ocosingo. Secretaria de Desarrollo Social. Retrieved from http://www.microrregiones.gob.mx/catloc/LocdeMun.aspx?ent=07&mun=059

Semitsu, J. P. (2011). From Facebook to Mug Shot: How the Dearth of Social Networking Privacy Rights Revolutionized Online Government Surveillance Social Networking and the Law. Pace Law Review, 31, 291–381.

Shaffer, G. (2013). Lessons Learned from Grassroots Wireless Networks in Europe. In A. Abdelaal (Ed.), Social and Economic Effects of Community Wireless Networks and Infrastructures: (pp. 236–254). IGI Global. https://doi.org/10.4018/978-1-4666-2997-4

Shilton, K. (2018). Engaging Values Despite Neutrality: Challenges and Approaches to Values Reflection during the Design of Internet Infrastructure. Science, Technology, & Human Values, 43(2), 247–269. https://doi.org/10.1177/0162243917714869

Shilton, K., Koepfler, J. A., & Fleischmann, K. R. (2014). How to See Values in Social Computing: Methods for Studying Values Dimensions. In Proceedings of the 17th ACM Conference on Computer Supported Cooperative Work & Social Computing (pp. 426– 435). Baltimore, MD, U.S.

Siddiqui, M. S., Montero, D., Serral-Gracià, R., Masip-Bruin, X., & Yannuzzi, M. (2015). A survey on the recent efforts of the Internet Standardization Body for securing inter- domain routing. Computer Networks, 80, 1–26. https://doi.org/10.1016/j.comnet.2015.01.017

Sollins, K. (1992, July). RFC 1350 - The TFTP Protocol (Revision 2). Internet Engineering Task Force (IEFT). Retrieved from https://tools.ietf.org/html/rfc1350

329

Spivak, G. C. (1994). Can the Subaltern Speak? In P. Williams & L. Chrisman (Eds.), Colonial Discourse and Post-Colonial Theory: A Reader (1st edition). New York: Columbia University Press.

Spurgeon, C. E. (2000). Ethernet: The Definitive Guide. O’Reilly Media, Inc. Retrieved from https://www.oreilly.com/library/view/ethernet-the- definitive/1565926609/ch01.html#ftn.ch01-FTNOTE-1

Star, S. L. (1999). The Ethnography of Infrastructure. American Behavioral Scientist, 43(3), 377–391. https://doi.org/10.1177/00027649921955326

Star, S. L. (2010). This is Not a Boundary Object: Reflections on the Origin of a Concept. Science, Technology, & Human Values, 35(5), 601–617. https://doi.org/10.1177/0162243910377624

Star, S. L. (2015). Living Grounded Theory: Cognitive and Emotional Forms of Pragmatism. In G. C. Bowker, S. Timmermans, A. E. Clarke, & E. Balka (Eds.), Boundary objects and beyond: working with Leigh Star (pp. 121–141). Cambridge, Massachusetts: The MIT Press.

Star, S. L., & Bowker, G. C. (2010). How to Infrastructure. In Handbook of New Media: Social Shaping and Social Consequences of ICTs, Updated Student Edition (pp. 230–245). London: SAGE Publications Ltd. https://doi.org/10.4135/9781446211304

Star, S. L., & Griesemer, J. R. (2015). Institutional Ecology, “Translations,” and Boundary Objects: Amateurs and Professionals in Berkeley’s Museum of Vertebrate Zoology, 1907-1939. In G. C. Bowker, S. Timmermans, A. E. Clarke, & E. Balka (Eds.), Boundary objects and beyond: working with Leigh Star. Cambridge, Massachusetts: The MIT Press.

Star, S. L., & Ruhleder, K. (2015). Steps Toward an Ecology of Infrastructure: Design and Access for Large Information Spaces. In Boundary objects and beyond: working with Leigh Star (1st edition, pp. 377–415). Cambridge, MA: The MIT Press.

Summerton, J. (1999). Power plays: the politics of interlinking systems. In O. Coutard (Ed.), The Governance of Large Technical Systems (pp. 93–113). Florence, UNITED STATES:

330

Routledge. Retrieved from http://ebookcentral.proquest.com/lib/aul/detail.action?docID=165427

Sundberg, J. (2014). Decolonizing posthumanist geographies. Cultural Geographies; London, 21(1), 33–47. http://dx.doi.org.proxyau.wrlc.org/10.1177/1474474013486067

Sweeney, L. (2013). Discrimination in Online Ad Delivery. ArXiv:1301.6822 [Cs]. Retrieved from http://arxiv.org/abs/1301.6822

TEDx Talks. (2015). Big Brother in einer digitalisierten Gesellschaft | Klaus Landefeld | TEDxRheinMainSalon. Frankfurt, Germany. Retrieved from https://www.youtube.com/watch?v=WqE5vxq79jY

Telmex. (2016, June 30). Condiciones del Servicio Infinitum Hogar para Mercado Residencial y Comercial (Masivo). Retrieved from https://downloads.telmex.com/pdf/cond-serv- infinitum-hogar.pdf

Telmex. (n.d.). Retrieved from http://blog.telmex.com/?s=fibra+optica

Testart, C. (2018). Reviewing a Historical Internet Vulnerability: Why Isn’t BGP More Secure and What Can We Do About it? (pp. 1–36). Presented at the TPRC 46: The 46th Research Conference on Communication, Information and Internet Policy 2018, Washington, DC: Social Science Research Network. Retrieved from https://papers.ssrn.com/abstract=3141666

Thussu, D. K. (2018). International Communication: Continuity and Change (3rd Edition). Bloomsbury Publishing.

U. S. Department of Homeland Security. (2011). Analyst’s Desktop Binder. U. S. Department of Homeland Security. Retrieved from https://epic.org/foia/epic-v-dhs-media- monitoring/Analyst-Desktop-Binder-REDACTED.pdf

UNESCO. (1945, November 16). UNESCO Constitution. UNESCO. Retrieved from http://portal.unesco.org/en/ev.php- URL_ID=15244&URL_DO=DO_TOPIC&URL_SECTION=201.html

331

UNESCO. (1980). Many voices, one world: Towards a new more just and more efficient world information and communication order (Reprint). London, New York: Kogan Page. Retrieved from http://www.un-documents.net/macbride-report.pdf

United Nations. (2018). WORLD ECONOMIC SITUATION AND PROSPECTS. NEW YORK: UNITED NATIONS PUBLICATION.

United Nations Conference on Trade and Development. (n.d.). UNCTADstat - Classifications. Retrieved January 19, 2019, from http://unctadstat.unctad.org/EN/Classifications.html

Vega Estrada, S. de la, Téllez Vázquez, Y., & López Ramírez, J. (2012). Índice de marginación por localidad 2010 (1. ed). México, D.F: Consejo Nacional de Población.

Vicentin, D. J. (2016). A reticulação da banda larga móvel: definindo padrões, informando a rede (Doctoral Dissertation in Sociology). Universidade Estadual de Campinas, Campinas, São Paulo. Retrieved from http://repositorio.unicamp.br/jspui/handle/REPOSIP/320968 von Burg, U. (2001). The Triumph of Ethernet: Technological Communities and the Battle for the LAN Standard (1st edition). Stanford, California: Stanford University Press.

Weiser, P. (2009). The Future of Internet Regulation (SSRN Scholarly Paper No. ID 1344757). Rochester, NY: Social Science Research Network. Retrieved from https://papers.ssrn.com/abstract=1344757

Weller, D., & Woodcock, B. (2013). Internet Traffic Exchange: Market Developments and Policy Challenges (OECD Digital Economy Paper No. 207). OECD Publishing. Retrieved from https://econpapers.repec.org/paper/oecstiaab/207-en.htm

Wilson, A. (2009). Computer Gap: The Soviet Union’s Missed Revolution and Its Implications for Russian Technology Policy. Problems of Post-Communism, 56(4), 41–51. https://doi.org/10.2753/PPC1075-8216560404

Winner, L. (1978). Autonomous Technology: Technics-out-of-control as a Theme in Political Thought. MIT Press.

Winner, L. (1980). Do Artifacts Have Politics? Daedalus, 109(1), 121–136. 332

Wolvers, A., Tappe, O., Salverda, T., & Schwarz, T. (2015). Concepts of the Global South. Cologne, Germany: Global South Studies Center Cologne. Retrieved from http://www.uni-koeln.de/

World Bank. (2018). World Bank Country and Lending Groups. Retrieved January 19, 2019, from https://datahelpdesk.worldbank.org/knowledgebase/articles/906519-world-bank- country-and-lending-groups

Yu, M., Jiang, W., Li, H., & Stoica, I. (2012). Tradeoffs in CDN Designs for Throughput Oriented Traffic. In Proceedings of the 8th International Conference on Emerging Networking Experiments and Technologies (pp. 145–156). New York, NY, USA: ACM. https://doi.org/10.1145/2413176.2413194

Zehle, S. (2012). New World Information and Communication Order. In The Wiley-Blackwell Encyclopedia of Globalization (Vol. 3, pp. 1529–1531). Chichester, West Sussex; Malden, MA: Wiley-Blackwell.

Ziewitz, M., & Pentzold, C. (2014). In search of internet governance: Performing order in digitally networked environments. New Media & Society, 16(2), 306–322. https://doi.org/10.1177/1461444813480118

Zittrain, J. (2008). The Future of the Internet and How to Stop It (1st Edition). New Haven: Yale University Press. Retrieved from https://dash.harvard.edu/handle/1/4455262

Zuboff, S. (2015). Big other: surveillance capitalism and the prospects of an information civilization. Journal of Information Technology; London, 30(1), 75–89. http://dx.doi.org.proxyau.wrlc.org/10.1057/jit.2015.5

Zuboff, S. (2019). The Age of Surveillance Capitalism: The Fight for a Human Future at the New Frontier of Power (1st Edition). New York: PublicAffairs.

333

ANNEX 1

OSI VS. TCP/IP LAYERS

Table 1A. Open Systems Interconnection (OSI) Reference Model and TCP/IP Protocol Stack

OSI Layer OSI Layer TCP/ IP Description Number Name Layer Name

Consists of standard communication services and 7 Application applications that everyone can use.

Ensures that information is delivered to the 6 Presentation Application receiving machine in a form that the machine can understand.

Manages the connections and terminations 5 Session between cooperating computers.

Manages the transfer of data. Also assures that 4 Transport Transport the received data are identical to the transmitted data.

Manages data addressing and delivery between 3 Network Internet networks.

Handles the transfer of data across the network 2 Data Link Data link media.

Physical Defines the characteristics of the network 1 Physical network hardware.

Source: Reproduction with adaptations from Oracle’s System Administration Guide: IP Services (Oracle, n.d.-a, n.d.-b).

334

ANNEX 2

INTERNET TRAFFIC PATTERNS

Figure 1A. Changes in Internet Traffic Patterns Representations. Source: Reproduction from Labovitz et al. (2010) p. 78.

335

APPENDIX A

THE BASICS ON INTERNET PROTOCOLS AND ETHERNET PHYSICAL NETWORKS

Communication through the internet occurs based on many different protocols, which can be understood as agreed-upon conventions (Cerf & Kahn, 1974, p. 1), or internet language, as already simply put.65 “They are the blueprints that enable technical interoperability among heterogeneous technology products” (DeNardis, 2009, p. 6), and in their capacity, they are

“political,” given their impact on the way that information will flow and their social and economic consequences for global communication. They embed the values of standards designers and their design serves as an alternative form of public policy (DeNardis, 2009).

By understanding protocols and other technologies as conventions, authors have investigated the history of their standardization and establishments, including the Internet

Protocol (DeNardis, 2009), mobile broadband protocols (Vicentin, 2016), and the short message service (SMS) (Acker, 2014), shedding light on the politics of current communication infrastructures. On the other hand, engineers involved with the deployment of founding internet protocols have also highlighted decisions behind their design (Clark, 1988; 2018) and the values embedded in some standardization processes in the very beginning of the internet, such as

“robustness in the presence of communication unreliability and availability in the presence of congestion” (Postel, 1981b, p. 1)—high priorities in military communication networks.

While the significance of protocols that mediate peoples’ communications has been highlighted in scholarship, in everyday life, it tends to be invisible.

65 A repository of the internet underlying protocols can be consulted here: https://www.rfc-editor.org/standards. 336

1. Basic Protocols in the Internet Design

The Internet Protocol (IP) serves as the basis for any other protocol on the internet. Its design supports the internet founding architecture given that it offers “provision for addressing, type-of-service specification, fragmentation and reassembly, and security information” (Braden,

1989, p. 10). In the founding document of the Internet Protocol, John Postel, an internet pioneer, explains it as follows:

The internet protocol provides for transmitting blocks of data called datagrams [data packets] from sources to destinations, where sources and destinations are hosts identified by fixed length addresses. The internet protocol also provides for fragmentation and reassembly of long datagrams, if necessary, for transmission through "small packet" networks. (Postel, 1981a, p. 1)

Some of the functionalities described above are executed by the junction of TCP/IP

(Transmission Control Protocol/Internet Protocol), which used to be a unique protocol, but by working separately, allow different data transport services to run concurrently on the internet

(Clark, 1988, p. 4).66 TCP runs over IP: it is part of the Transport Layer, known as layer-4, while

IP is part of the Network or Internet Layer known as layer-3, where the networks’ routing decisions conceptually take place. Both protocols complement each other as David Clark, another U.S internet pioneer, explains:

The IP layer of the architecture contains the address information which specifies the destination host to which the datagram [data packet] is being sent. (…) IP performs only a very simple dispatching once the datagram has arrived at the target host, it dispatches it to a particular protocol. It is the responsibility of that protocol handler, for example TCP,

66 There are other Internet Protocol auxiliaries--protocols that support it with error and other control tasks (Internet Control Message Protocol – ICMP), instructions for routers and hosts (Internet Group Management Protocol – IGMP) (F. Baker, 1995), and executing what are considered more primary tasks in the transport layer, than the ones that TCP execute, such as matching host names and their respective IP addresses to enable packets traffic; and underlying basic transfer files protocols such as the Trivial File Transfer Protocol (TFTP), which has fewer features than the more popular File Transfer Protocol (FTP) (Postel, 1979, 1980; Sollins, 1992). 337

to finish dispatching the datagram to the particular connection for which it is destined (Clark, 1982, p. 9).67

The internet layers are useful to further understand that most internet exchange points work below the aforementioned layers, in layer-2, which means that they impact network routing decisions, but are not the primary actors responsible for them. Because IXPs shorten the distance between networks, their presence ends up attracting the paths selected to forward data packets; also, when autonomous systems decide to become an IXP member for any reason, that amplify their routing options and influence the outputs that their routers’ algorithms, programmed by network engineers, will present. Thus, the internet layers are abstractions used to explain the internet (see Annex 1), however they are by no means rigorously defined borders. Instead, as already explained elsewhere in terms of underlying protocols: “Protocols in different layers interact in complex and sometimes subtle ways, and particular functions often involve multiple layers. There are many design choices in an implementation, many of which involve creative

"breaking" of strict layering” (Braden, 1989, p. 15). Thus, it is more useful to think of layers as elements of an organic environment, and in this sense, when IXP spokesmen argue that they are not responsible for forwarding packets because this occurs in another layer where autonomous systems’ routers are, one can understand their protective argument without disregarding the influence that IXPs have on that.

67 Peter Denning and Craig Martell make distinctions between hosts and network. The latter “is the set of routers and links that move packets” while the hosts “are systems attached to the network through standard interfaces” (Denning & Martell, 2015, p. 265). Simply put, a host is the origin and destination of data packets. It can be, for instance, a user’s workstation or a mainframe with large differences in speed and functions (Braden, 1989, p. 7). Host is “the ultimate consumer of communication services [, which] generally executes application programs on behalf of user(s), employing network and/or Internet communication services in support of this function” (Braden, 1989, p. 6). 338

2. Ethernet: The Local Network

For data packets to travel from one point to another, physical connections between computers are necessary. By briefly discussing some physical network design options in this section, the assumption is that the physical IXP design will become clearer.68

Local Area Network (LAN) is the name given to networks within a restricted geographic area, such as rooms or buildings. When LANs are based in different buildings but need to share resources, they can form a Virtual Local Area Network (VLAN), which is the case of giant IXPs.

One of the most widely used LAN standards in the world, and the common framework used in

IXP LANs, is Ethernet, which is defined as “a system for local communication among computing stations” (Metcalfe & Boggs, 1976, p. 396), and whose very goal is to enable sharing of information among computers with simplicity and low cost (Abbate, 1999, p. 187). The prefix

“Ether” comes from “luminiferous ether,” a substance that was once believed to be required for the propagation of electromagnetic waves. In the 19th century, such belief was refuted, being however revived with the memo that originated “The Ether Network,” written by Bob Metcalfe in 1973 at the Xerox Palo Alto Research Center (PARC) (Motta, n.d.; Schley, 2017; Spurgeon,

2000). The ether represents, thus, the medium, and more precisely, the materiality of a network through which data packets will traverse, referring to basically the physical layer (layer-1) and the layer-2 switch. Bob Metcalfe and David Boggs explain that:

An Ethernet's shared communication facility, its Ether, is a passive broadcast medium with no central control. Coordination of access to the Ether for packet broadcasts is distributed among the contending transmitting stations using controlled statistical

68 It is important to note that although this section emphasizes the physical aspects of a local network, networks are much logical as physical, and their configuration is software defined. 339

arbitration. Switching of packets to their destinations on the Ether is distributed among the receiving stations using packet address recognition (Metcalfe & Boggs, 1976, p. 395).

It is useful to notice some of the values embedded in the Ethernet’s description above such as passivity, distributed control using mediated statistical and identifier recognition to transmit packets. Passivity aimed to avoid that a failure in an active and localized element of the network would affect the whole system, and is thus associated with resilience. The non- centralized control aimed to avoid bottlenecks and increase reliability. Finally, being able to read identifiers is intrinsic to the packets transportation task.

Regarding the Ethernet topology, it is supposed that a unique path should exist between any two end-points within a LAN network to avoid interferences. In this way, the network “has central interconnection through the Ether and distributed control among its stations” (Metcalfe &

Boggs, 1976, p. 397). Decisions are thus left to computer hosts or stations. Moreover, with an address in the header, the authors explain that in its origin, “A packet placed on the Ether eventually propagates to all stations. Any station can copy a packet from the Ether into its local memory, but normally only an active destination station matching its address in the packet's header will do so as the packet passes” (Metcalfe & Boggs, 1976, p. 397). On this topic, Peter

Dordal explains that this is still valid for today’s Ethernet, that packets are seen even if they are not directed to the devices’ specific addresses: “Packets addressed to other physical addresses will be seen by the [Ethernet] card, but ignored (by default)” (Dordal, 2018, p. 48).

Clearly, “by default” implies the existence of good practices shared among network operators, however there are possibilities of devices, through their cards’ software, to be manipulated and being put in a “promiscuous mode” for accepting, instead of ignoring, all the packets received. As Peter Dordal points out:

340

This mode was originally intended for diagnostic purposes but became best known for the security breach it opens: it was once not unusual to find a host with network board in promiscuous mode and with a process collecting the first 100 bytes (presumably including user id and password) of every telnet connection (Dordal, 2018, p. 48).69

Interestingly, “promiscuous” implies a values judgment and indicates a break of established norms, or good practices, in the name of security, even though this also inserts more insecurity into the system.

To extend the network, it was expected that any new computer “tap[pped] into the Ether at the nearest convenient point” (Metcalfe & Boggs, 1976, p. 397) to join it. As Peter Dordal further explains, showing the current deployment from the original Ethernet design, what used to be called “tap,” or mid-cable attachment, now is done with active devices:

In the original thick-coax cabling, connections were made via taps, often literally drilled into the coax central conductor. Thin coax allowed the use of T-connectors to attach hosts. Twisted-pair does not allow mid-cable attachment; it is only used for point-to-point links between hosts, switches and hubs. Mid-cable attachment, however, was always simply a way of avoiding the need for active devices like hubs and switches.” (Dordal, 2018, p. 47).70

Based on the topology described, since its origins, the Ethernet enabled a system where network machines are responsible for accessing or ignoring data traversing the shared medium.

Computers are also responsible for seeking quality and reliable communication processes, as the network itself is subject to transmission errors, in order to prioritize economy and a general, instead of specific, reliability level.

69 Telnet is a protocol used to remotely access computer hosts from another computer terminal (Davidson et al., 1977). 70 According to Peter Dordal, hubs are connecting devices similar to switches with some specific differences, as processing bits individually instead of integral packets as switches do. And whereas, according to the author, hubs have been increasingly replaced by switches over the years, “There is still a role for hubs today when one wants to monitor the Ethernet signal from A to B (e.g. for intrusion detection analysis), although some switches now also support a form of monitoring” (Dordal, 2018, p. 47). 341

2.1. Values in Ethernet Design

The Ethernet was standardized first by a consortium of companies, DIX, named after the

Digital Equipment Corporation, Intel Corporation and Xerox Corporation. The first specification was launched in 1980 and the second in 1982 (Dordal, 2018).71 Until today, it is not uncommon to find documents with references to “DIX Ethernet” (NIC.br & IX.br, 2017; Spurgeon, 2000), even if since 1985, the Ethernet has been a standard maintained by the Institute of Electrical and

Electronics Engineers (IEEE) instead. In fact, authors recognize that some DIX specifications are hard to change, even in face of new IEEE definitions, as occurred with tentative changes of the fields contained in the Ethernet data packets (Dordal, 2018, p. 47).

For its permanence and significance for today’s internet, the DIX specification guidelines are useful to examine the values embedded in the “goals” and “non-goals” of this network standard in its very beginning. Among the listed goals are “simplicity,” “low cost,”

“compatibility” (meaning inter-operationability), “high speed,” “low delay” and “stability”

(Digital Equipment Corporation et al., 1982, p. 5). There are also two other principles that aim to provide machines some level of equality and independence: “fairness” aims to guarantee that

“All stations should have equal access to the network when averaged over time.” Progress,” in turn, supports that “No single station operating in accordance with the protocol should be able to prevent the progress of other stations” (Digital Equipment Corporation et al., 1982, p. 5).

71 For a historical account on the establishment of Ethernet see von Burg (2001). The author shows that in the 1980s, Ethernet was competing with the IBM’s Token Ring standard, and both were under discussions at the IEEE. In his thesis, the Ethernet’s open technologies and late development contributed to its dominance. 342

With regard to security, error control was originally “limited to detection of bit errors in the physical channel, and the detection and recovery from [packets] collisions” (Digital

Equipment Corporation et al., 1982, p. 6). Also, no encryption was provided at this layer and “no attempt to protect the network from a malicious user at the data link [layer-2] level” existed

(Digital Equipment Corporation et al., 1982, p. 6). Clearly, there was a great amount of trust embedded in the Ethernet design, which would be enforced by standards that, in context, work as norms to be followed. Currently, the Ethernet standards are issued by the Institute of Electrical and Electronics Engineers (IEEE) through the 802.3 Ethernet Working Group. Specifications are dynamic and the solutions aim to respond to the increasing traffic that characterizes the current capacities of a Gigabit Ethernet.72

72 Archives for the documents issued by the 802.3 Ethernet Working Group of the Institute of Electrical and Electronics Engineers (IEEE) are kept at http://www.ieee802.org/3/, while the latest Ethernet standard was published in 2016 (IEEE, 2016). 343

APPENDIX B

SWITCHES AND ROUTERS COMMUNICATION AND NETWORKS PATH

PRIOTIZATION

1. The Role of Switches and Routers in a Packet Switching Internet

The internet is known as a “packet switching” network because information traversing the medium is broken into many pieces and enveloped into Internet Protocol packets with unique destination addresses. Such addresses orient the path that packets will take, crossing one or more networks until they arrive at their destination host. This mesh configuration is opposed to what characterized the telecommunications networks, which were relatively more centralized and hierarchical networks.

Peter Denning and Craig Martell localize packet switching as an example of the

“multiplexing principle,” characteristic of networks. According to the authors, “Multiplexing means to divide a resource into chunks and then allocate the chunks to various individuals”

(Denning & Martell, 2015, p. 220, emphasis in the original), where “individuals” may be understood as packets in this context. The authors explain that such principle was beneficial for the internet because of three reasons: first, it allowed traffic of different kinds to occur simultaneously, second, because it brought advantages in relation to the telephone system, and third, because it enabled the possibility of circumventing system failures, so the packets have different routes to follow. The consequence of packet-switching and the multiplexing principle is shaping the internet as a system where control and decision are distributed and responsibility for the network to work is shared. 344

Importantly, in this design, each recipient network defines the next paths for the packets received—following the distributed architecture designed in the 1960s, when each network was conceived to be autonomous and to set its internal rules (Abbate, 1999). The TCP finally reassembles the packets when they arrive in their destination using numbers in the packets that identify their sequence (Cerf & Kahn, 1974, p. 6).

To highlight the significance here, the interconnection arrangements, Vint Cerf and Bob

Kahn, in one of the founding papers of the TCP/IP protocols, advance the efforts of coordination to establish communication between autonomous networks. Focusing on the network borders to guarantee network autonomy, they explain:

To allow networks under different ownership to interconnect, some accounting will undoubtedly be needed for traffic that passes across the interface… the interconnection must preserve intact the internal operation of each individual network…. It is thus apparent that the interface between networks must play a central role in the development of any network interconnection strategy. We give a special name to this interface that performs these functions and call it a GATEWAY. (Cerf & Kahn, 1974, p. 2, emphasis added)

Nowadays, gateways are known as routers (Baker, 1995; Denning & Martell, 2015)— hardware with a key role in network interconnection for building their route databases, also known as routing tables, and running algorithms that store and forward packets.

Comprehensively, a router supports the internet protocols that define its architecture and has the following functions: enveloping and un-enveloping data packets with the network identifiers

(e.g. headers); translating IP addresses into hardware addresses inside a network when necessary; receiving and dropping data packets according to rules; supporting network errors monitoring and network management; breaking data packets into parts if needed; forwarding data packets based on information in its routing table; being compatible with protocols that support reachability algorithms intra and inter-networks (known as interior and exterior gateway 345

protocols, respectively) to find the specific packets’ destination hosts (Baker, 1995, pp. 27-28). It is also expected to send the fragments of packets in order and to reassemble them when necessary (Baker, 1995, pp. 45-46).

In accordance with its functionalities, a router is supposed to be agnostic in relation to data packet identifiers, which means that these devices “are designed to be stateless, forwarding each IP datagram [packet] independently of other datagrams” (Baker, 1995, p. 30; Braden, 1989, p. 7). In this design, “All state information required for end-to-end flow control and reliability is implemented in the hosts, in the transport layer or in application programs” (Baker, 1995, p. 30;

Braden, 1989, p. 7). Nevertheless, routers are supposed to read IP information to forward and to reassemble packets that were fragmented. Regarding the reassemble task, a router needs to access, for example, an internet identification field (ID), the source and destination address, among other information (Postel, 1981a, p. 24). Thus, routers’ agnostic role should not be confounded with no capacity to read data.

Furthermore, a router may apply kinds of filters, defining packets that should and should not be forwarded based upon “message definitions,” which “[specify] the source and destination network prefix, and may include other identifying information such as IP Protocol Type or TCP port number” (Baker, 1995, p. 98, emphasis added). Given that, it is important to highlight that the fact that devices and network operators are, by norms, agnostic about what is traversing the network does not mean that they do not have access to identifying information. Above all, it means that they are supposed to use such information appropriately and only when their functions require it.

346

Routers in the path inscribe their identification (ID) number into the IP packet header.

This can be an IP address or another ID number, with which it is possible to know, for example, with which network a router is associated (Baker, 1995, p. 42). Routers’ identification is an important piece of information of the Autonomous System Path (AS_PATH), an attribute of the router communication protocol, which will be explained in the next section, that allows it to know the path that data packets take from origin to destination and to create metrics based on that.

Comparing autonomous systems’ routers functions with IXP switch functions, which work together in an IXP environment, it is important at this point to differentiate their roles and decision-making capacities. Routers and switches are different devices, but as routers can function as switches sometimes, it is better to differentiate them in terms of their functions and not in terms of devices. Switches are used to establish communication between autonomous systems part of the same network (e.g. the IXP network), and forward packets among them.

Working conceptually at layer-2, switches allow less control and configuration than routers, which, in turn, are positioned at layer-3 and are used to communicate between a variety of networks that are not necessarily alike (Cisco, 2014).

To communicate, a “switch reads in and forwards an entire packet as a unit, and the destination address is consulted to determine to where the packet is forwarded” (Dordal, 2018, p.

47). Unlike routers, switches do not use the IP addresses to identify data packets’ destinations and forward the packets to a host computer. Instead, they use hardware addresses, also known as

MAC (Media Access Control) addresses. In this case, switches consult a forwarding table to match the hardware address to the IP address that corresponds to the destination host, and only

347

then forward the packet. For a better understanding, in this case, the role of a layer-2 switch is like “(…) to deliver mail using social-security numbers [SSN] addresses, where each postal worker is provided with a large catalog listing each person’s SSN together with their physical location” (Dordal, 2018, p. 23).

Nevertheless, from Network Layer (layer-3)—where routers are situated—to Data Link

Layer (layer-2)—where IXP switches are situated, for any packet traversing the layers it is necessary to receive the following: “(1) The IP packet (…) (2) The length of the IP packet (…)

(3) The destination physical interface (…) (4) The next hop IP address (5) The Link Layer priority value (…) [and] if the packet to be transmitted causes a Link Layer precedence-related error” (Baker, 1995, p. 33). This happens because everything is running on a TCP/IP internet and data traversing layers is still IP data packets with its characteristic identifiers. The significance of that is that even if the layer in which IXPs are conceptually situated are devoid of many decision tasks, switches have access to the whole packet’s information that is normatively supposed to be ignored.

Finally, considering the packet switching process, one can ask if different data packets that form a single piece of information can travel through separate paths towards its destination.

Network operators say that this is unlikely to happen nowadays. The first reason is that, traveling through multiple paths, the latency generated will be different, so packets will arrive unordered, which is not desirable in a TCIP/IP designed internet. Second, to cope with high volumes of data, hardware has also been adapted. Routers do not look up a destination for every single packet anymore, according to Klaus Landefeld, the engineer responsible for the infrastructure of DE-

CIX, explains in an interview with the author. Instead, routers look up the packet destination just

348

once, and download the flow forwarding information into the network. When the next packet arrives, the network itself forwards it to the outbound port, without sending it to a central process unit (CPU). According to him:

Only that is efficient. That is standard. That is normal. Every hardware is built this way. Otherwise, the central process would not be able to handle terabits of data. The number of packets per second is so high, that the CPU would not be able to cope. So you need to have that direct forwarding. Because of that, the actual idea of traversing separate paths through the network, while in theory it is still possible, it is not a very good idea to do it in that way.

Therefore, if the path decision algorithms define that a data packet should traverse an

IXP, the whole piece of that information is going to traverses it as well—a surprising additional point of centralization in the information circulation infrastructure of the internet to remember us that “technological trajectories are not easy to grasp and that they do not go straight—no more than do the beings responsible for the establishment of chains of reference” (Latour, 2013, p.

216, emphasis in the original). important design characteristic of the network.

2. ISPs Communication and Paths Prioritization

While routers work at the Internet Protocol layer and support the functioning of other related protocols, their communication with routers of different networks, known as inter-domain communication, are based on the Border Gateway Protocol (BGP). BGP can be understood as the language spoken by routers that allows them to advertise or announce the paths to reach networks and to learn such information from neighbor routers. It works as an exterior routing protocol—as opposed to an interior routing protocol that provides mostly intra-network, or intra-

349

domain routing information.73 According to the Request for Comments that introduces the BGP in use nowadays (version 4), “The primary function of a BGP speaking system is to exchange network reachability information with other BGP systems. This network reachability information includes information on the list of Autonomous Systems (ASes) that reachability information traverses” (Rekhter et al., 2006, p. 3, emphasis added).74

BGP allows for the creation of dynamic route databases that are constantly updated

(Mathew, 2016). Additionally, BGP also defines how routers will learn routes and specifies mechanisms to the construction of routes (Mathew, 2016). The primary task of an autonomous system is to advertise its own prefixes or addresses and the addresses to reach its clients, and such behavior constitutes the boundary of peering relationships. By market practice, Internet

Service Providers do not advertise routes that they learn from their transit providers, known as upstream routes, or from its peers at an IXP; such information must be used to improve their own internet reachability, connectivity and consequent competitiveness before other ISPs. The significance of this behavior is that by advertising just their own routes and clients’ routes, autonomous systems move the internet economy through both cooperation and competition.

For a better understanding, the autonomous system’s router (or routers) announces its

(their) sets of IP addresses in the format of prefixes, which are linked to hosts’ addresses, or destination addresses, as the notation below shows:

73 As an exception, there are situations where an exterior routing protocol as BGP is disregarded and communication among interior routing protocols is used instead (Baker, 1995, p. 110). 74 The word router comes from the role of building the routing table (Baker, 1995, p. 19), and interestingly implies action to that technological device. Further, the verbs learn, advertise, announce and speak emphasized in this paragraph are commonly used in the engineering scholarship as well as in network engineers’ interviews with the author applied to technologies. In reproducing such use, I am guided by STS literature that considers the agency of technological artifacts and humans symmetrically. 350

“IP-address ::= { , }” (Baker, 1995, p. 24)75

As autonomous systems advertise their own prefixes, they become the origin networks for those prefixes, then autonomous systems’ routers advertise the BGP paths to reach them allowing that, dynamically, other routers learn with that. For instance, if network A received data packets from network D through networks B and C, network A now knows that when necessary to reach D, it can follow the path B – C – D (Roberts et al., 2011). This does not mean that a learned path will necessarily be followed in the opposite direction, though. Indeed, the internet is commonly routed asymmetrically, which means that the inbound path—that comes from other networks—will follow a different route than the outbound path—the one that leaves a network

(DE-CIX, n.d.-b).

In this regard, algorithms define the steps that routers will use to select the addresses and forward the data packets received to the best route by examining routers’ Forwarding

Information Base (FIB) (Baker, 1995, p. 74). In their most basic form, routers should aim to find the straight course to a destination, as idealized in the pioneering algorithms by Edsger Dijkstra as early as 1959 (Denning & Martell, 2015, p. 225). Nevertheless, routing algorithms reflect the complexity of the internet currently. Attributes are defined based on the policies of the autonomous system operators, which reflect economic interests and commercial agreements that aim at optimizing their costs. Network operators use prioritization rules and define preference values to routes, ranking them between the most and least preferred routes. This procedure has been referred to as “administrative preferences” (Baker, 1995, p. 77).

75 Some prefixes are more specific, with fewer hosts’ addresses linked to them, while others are less specific for having more hosts’ addresses associated. By norm, routers prioritize specific prefixes over others to forward its traffic (Baker, 1995, p. 24; Testart, 2018). 351

Due to the business decisions that are embedded in the BGP configuration, BGP operators have defined it as a protocol based on “backroom politics,” as Tiago Felipe Gonçalves, a network engineer and IXP operator in Brazil, described it, highlighting why it becomes a complex protocol. In a similar direction, authors suggest that “BGP is not a very complicated protocol, but the way it is operated in practice, that is, allowing flexible policies while maintaining global scalability, makes it intricate” (Siddiqui et al., 2015, p. 2).

BGP is not centrally coordinated. Instead, it works based on unique standards and the coordination of individuals who operate the networks, through relationships of trust and a

“distributed governance” (Mathew, 2016; 2014). As information science researchers note, such governance goes beyond the market logics and hierarchical arrangements. It relies on trust and collaboration, while it is also mediated by corporations and nation-state interests (Mathew,

2014). In the same direction, but adopting an economic sociologic framework, Uta Meier-Hahn defends that the interconnection realm is ruled by factors that are not totally captured by the traditional market analysis: global coordination, trust and mitigation of uncertainties are important characteristics that she observes on the ground (Meier-Hahn, 2014b).

BGP security is recognizably an enormous problem in the engineering and policy communities and in scholarship. Analyzing the deployment of the current BGP version over the years through its technical documents, Cecilia Testart argues that the BGP debate stands between two ends: developers’ efforts to prioritize BGP availability in order to guarantee an effective flow of information online, and the need to improve BGP security; yet both goals may defy each other (Testart, 2018). Such a trade-off is characteristic of the BGP design. Possible disruptions that can occur in BGP are based on fake routing information: for instance, by seeming to

352

announce a prefix that is not its own, an autonomous system can wrongly “steal” traffic from another network, making the target network unreachable and subjecting the packets forwarded to it to malicious use (Murphy, 2006, p. 4).76

In summary, BGP-distributed governance does not count on validation mechanisms that could prevent this kind of event. The protocol is considered vulnerable in terms of guaranteeing authenticity and integrity of autonomous systems and routes advertised (Siddiqui et al., 2015).

As Cecilia Testart notes, this scenario is very different from the highly hierarchized governance structure that runs IP addresses (Testart, 2018).77 In this realm, the Internet Assigned Numbers

Authority (IANA) allocates blocks of IP addresses to continent-based Regional Internet

Registries (RIRs) (e.g. Latin America & Caribbean Network Information Center [LACNIC];

Réseaux IP Européens Network Coordination Centre [RIPE NCC]). Such allocation, which is also known as delegation for allowing management and sub-delegations by RIRs in the regions, enables these organizations to allocate blocks of IP addresses to Local Internet Registries (LIRs), which in turn will assign such blocks to Internet Service Providers (ISPs)–where assigning implies that the organization receiving the resources will use them directly (Fuller & Li, 2006, p.

6). RIRs and LIRs follow their own policies and prices to allocate and assign IP address blocks.

There are cases in which RIR directly assigns address blocks to ISPs, for example. They may also regulate differently how assigned blocks can be managed, if they can be resold or not.

An important aspect of the internet registry address system explained above that directly impacts routers’ communication is the format of IP address blocks or prefixes. Prefixes are what

76 One of the most known cases is the traffic steal of YouTube in 2008 to a Pakistani network after the government of Pakistan tried to block that site locally, and as a result, a fake route has been disseminated globally (Kuerbis & Mueller, 2017; Mueller, 2010; RIPE NCC, 2008). 77 In fact, this hierarchical governance also applies to autonomous systems, which are also allocated. 353

routers announce in a BGP speaking session when communicating with another device, and, depending on prefix size, they will require more processing capacity in terms of memory and

CPU. For example, in the prefix 1.0.6.0/24, while “1.0.6.0” identifies the network, “/24” indicates that there are 256 IP addresses in this block, each of which may be a specific host to be reached. Prefixes range from /32, containing a unique IP address, to /0 containing 4.294.967.296 addresses (Fuller & Li, 2006, p. 7). According to Adilson Florentino, a network specialist interviewed by the author, the most common is to announce prefixes from /24 to /21, which means blocks from 256 to 2048 IP addresses, respectively. IXPs can also have their own norms: in Brazil, the IX.br accepts announcements of /8 to /24 only, to avoid blocks considered too big or too small being processed. In fact, as of January 2016, of 580.000 prefixes in a IPv4 routing table, 318,000, or approximately 55%, were /24 (NSRC, 2016, P. 10).

Nevertheless, technically, autonomous systems can subdivide their blocks and work with sub-blocks. They can, for instance, subdivide a /24 prefix, with 256 IP addresses into two blocks of /25, with 128 IP addresses each, in order to respond to their own policies or to negotiate one of the sub-blocks with clients that need such resources in markets with scarce IPs addresses that allow this kind of negotiation. This makes the control on the actual use of IP addresses worldwide a hard task.

354