Results of Second Batch of Selected Experiments

European Research 7th Framework Programme

Project title: Community Networks Testbed for the Future Internet.

Results of second batch of selected experiments

Deliverable number: D.4.7 Project Acronym: CONFINE Project Full Title: Community Networks Testbed for the Future Internet. Type of contract: Large-scale integrating project (IP) contract No: 288535 Project URL: http://confine-project.eu Editor: Bart Braem, iMinds Deliverable nature: Report (R) Dissemination level: Public (PU) Contractual Delivery Date: June 30, 2015 Actual Delivery Date September 30, 2015 Suggested Readers: Project partners Number of pages: 134 Keywords: WP4, open call, experimental research, community networks, testbed Authors: Carlos Rey-Moreno, Tafadzwa Mandava, Lwando Mdleleni, Renette Blignaut - University of the Western Cape Thomas Huhn,¨ Stefan Venz - Dai Labor Greta Byrum - New America Foundation Paul Fuxjaeger - FTW Monica Gariga, Narcis Vives - Itinerarium Andrea Detti - CNIT Claudio Pisa - Unidata Ahmed Abujoda, Panagiotis Papadimitriou - University of Hannover Arjuna Sathiaseelan - University of Cambridge Roger Pueyo Centelles - Routek Vassilis Chryssos, Giorgos Klisiaris - Sarantaporo.gr Peer review: Christoph Barz - FKIE Julia Niewiejska - FKIE Aaron L. Kaplan - Funkfeuer

Abstract

The CONFINE project studies community networks and has developed a testbed to allow experimentation with community networks. To evaluate the testbed and to stimulate broad adoption of community networks research, the project includes two open calls for participation. A second open call was launched in project year 4. Nine partners were selected to do research on and with community networks, supported by the project. This deliverable gives a summary of experiences and results from this second open call. In the COSMOS project (CrOwd-Shared Mesh netwOrk for universal internet Service), the University of Cambridge and the Leibniz Universitat¨ Hannover investigate the benefits of extending the coverage of any crowd-shared network (e.g. the PAWS network in Nottingham, UK) by connecting the home routers as a mesh. The Technische Universitat¨ Berlin implemented a cross-layer joint power and rate controller, Minstrel-Blues, which enables realising different power control approaches using today’s WiFi hardware. Routek created software, called the Network Characterisation Daemon or NCD, an interactive tool that provides users of community networks a means of monitoring, evaluating and fine-tune their network nodes. In the BTC (“Blessing of The Commons”) project, the Telecommunications Research Center Vi- enna has built a dedicated interference measurement testbed using software-radio front-ends as signal transmitters. Additionally, the Linux kernel driver code has been modified and selected user-space tools have been implemented to allow keeping track of the instantaneous channel load and spectral efficiency characteristics for every link in the network. Itinirarium and the Institute of Government and Public Policies of the Autonomous University of Barcelona, in the CitizenSqKm project, explored how a community reacts to a new platform where digital information is created, collected, guarded, processed and disseminated by citizens in a collective and structured effort. The New America Foundation has made a contextual analysis of community network sustainability. They have developed a broad-based framework for understanding socioeconomic factors that determine the success, sustainability and social impact of community networks. In The Icarus project, the University of the Western Cape has performed a case study on a community network in rural South Africa. The impact of connecting the community network to the Internet has been investigated and show how this is instrumental to develop and reinforce other social structures within the community. In CONFLATE (CONFINE extension towards OpenFlow experimentation: infrastructure, software and demonstrations), UNIDATA and the Consorzio Nazionale Interuniversitario per le Telecomuni- cazione have expanded the Community-Lab testbed in two dimensions: infrastructural and functional by deploying new Community-Lab research devices in the Ninux.org network in Rome and by allowing researchers to deploy OpenFlow experiments in Community-Lab. Finally, Sarantaporo.gr has expanded its community wireless network and interconnected islet villages via a backbone infrastructure. Furthermore, it has expanded to the nearest technical institute and through this to other community networks all over Europe. Contents

I. Technological experiments8

1. Cosmos - CrOwd-Shared Mesh netwOrk for universal Internet Service9 1.1. Introduction ...... 9 1.2. Background ...... 9 1.2.1. Software Deﬁned Crowd-Shared Wireless Mesh Networks ...... 10 1.2.2. Trafﬁc Redirection ...... 11 1.2.3. User Sharing Policies ...... 12 1.2.4. Implementation ...... 14 1.3. Experiment Description ...... 17 1.4. Test Setup and Results ...... 17 1.5. Conclusions ...... 20

2. MinstrelBlues - Joint Transmit Power and Rate Control on CONFINE Testbed 22 2.1. Introduction ...... 22 2.2. Background ...... 22 2.3. Experiment Description ...... 24 2.4. Test Setup and Results ...... 25 2.5. Main Achievements and Challenges ...... 29 2.6. Conclusions ...... 31

3. Reflection - Enhancing Reflection and Self-Determination in Community Net- working 32 3.1. Introduction ...... 32 3.1.1. Objectives and Vision ...... 33 3.2. Related work ...... 33 3.3. Social and Technological context ...... 34 3.3.1. Community Mesh Networks in the Guifi.net environment ...... 34 3.3.2. Quick Mesh Project (qMp) ...... 34 3.3.3. BatMan-eXperimental version 6 (BMX6) routing protocol ...... 35 3.4. The Network Characterisation Daemon (NCD): a characterisation and interaction tool for Community Mesh Networks (CMNs) ...... 35 3.4.1. General concepts ...... 35 3.4.2. The NCD: architecture, components and considerations ...... 36 3.4.3. NCD source code and OpenWrt package for dissemination ...... 39 3.5. Achievements and Challenges ...... 39 3.6. Achievements and challenges for Community Mesh Networks ...... 39 3.6.1. Achievements and challenges for commercial exploitation ...... 40 3.6.2. Conclusions ...... 40

1 Contents Contents

4. Blessing of The Commons: Improving Radio Resource Utilization Efficiency in Community Wireless Networks 41 4.1. Motivation ...... 41 4.2. Part I - Empirical Model of Interference in IEEE802.11 ...... 42 4.2.1. Introduction ...... 42 4.2.2. Related Work ...... 43 4.2.3. The Current NS-3 Interference Model ...... 44 4.2.4. Measurement Methodology ...... 46 4.2.5. Measurement Results ...... 48 4.2.6. Adaption of the NS-3 Interference Model ...... 50 4.2.7. Summary of Part I ...... 51 4.3. Part II - Resource Utilization Awareness in IEEE802.11 ...... 52 4.3.1. Introduction ...... 52 4.3.2. Related Work ...... 53 4.3.3. Channel Load Measurement ...... 53 4.3.4. Channel Load Broadcasting ...... 54 4.3.5. Modifications to the mac80211 Subsystem ...... 55 4.3.6. Channel Load in HORST ...... 55 4.3.7. Relation to License Assisted Access ...... 56 4.3.8. Spectral Efficiency Measurement ...... 57 4.3.9. Summary of Part II ...... 58 4.4. Concluding Remarks ...... 59

II. Social experiments 60

5. Citizen Square Kilometer 61 5.1. Introduction ...... 61 5.2. Background ...... 61 5.3. Test Setup and Results ...... 63 5.3.1. An extension of the guiﬁ.net community network ...... 63 5.3.2. Itinerarium’s geolocation platform ...... 64 5.3.3. Ethnography and participatory techniques to Adopt, Adapt, Create and Share 65 5.4. Main Achievements and Challenges ...... 71 5.5. Conclusions ...... 74

6. New America Foundation - Contextual Analysis of Community Network Sus- tainability 76 6.1. Introduction ...... 76 6.2. Background ...... 76 6.3. Experiment Description ...... 77 6.4. Test Setup and Results ...... 78 6.4.1. Test Method 1: Geospatial analysis ...... 78 6.4.2. Test Method 2: Regulatory, Policy, and Market Analysis ...... 82 6.4.3. Test Method 3: M-Lab Integration ...... 86 6.5. Main Achievements and Challenges ...... 89 6.6. Conclusions ...... 90

Deliverable D.4.7 2 Contents Contents

7. ICARUS - Impact of Community networks as Alternative infrastructure in remote and Underserved areaS 92 7.1. Introduction ...... 92 7.2. Background ...... 93 7.3. Experiment Description ...... 94 7.4. Test Setup and Results ...... 96 7.4.1. Study the impact of the communication network on the communication expenditure and patterns of both mobile and non-mobile users ...... 96 7.4.2. Collection and identiﬁcation of other non-expected effects of the community network and the Internet connectivity ...... 98 7.4.3. Analysis of the inﬂuence of the community network in the agency and aspirations of users ...... 99 7.4.4. Description of the business model used to sustain the community network and its services ...... 101 7.5. Conclusions ...... 103

III. Testbed expansion 104

8. CONFLATE - CONFINE extension towards OpenFlow experimentation: infrastructure, software and demonstrations 105 8.1. Introduction ...... 105 8.2. Background ...... 105 8.2.1. Community-Lab Testbed Expansion in Ninux.org ...... 105 8.2.2. OpenFlow Experimental Facility (OFX) ...... 106 8.3. Experiment Description ...... 110 8.4. Test Setup and Results ...... 111 8.4.1. Test Setup ...... 111 8.4.2. Results ...... 114 8.5. Main Achievements and Challenges ...... 114 8.6. Conclusions ...... 116

9. Sarantaporo.gr WiFi Networks 117 9.1. Introduction ...... 117 9.2. Background ...... 117 9.3. Test Setup and Results ...... 118 9.3.1. Technical aspect ...... 118 9.3.2. Social aspect ...... 119 9.4. Main achievements and challenges ...... 121 9.5. Conclusions ...... 124

10.Conclusions 125 10.1. Technological Experiments ...... 125 10.2. Social Experiments ...... 126 10.3. Testbed Expansion ...... 126

Deliverable D.4.7 3 List of Figures

1.1. Crowd-shared WMN for public Internet access...... 10 1.2. Software-defined WMN control plane overview...... 12 1.3. Flow redirections...... 13 1.4. Experimental setup for packet reordering evaluation...... 14 1.5. Packet reordering...... 14 1.6. SDWMN controller...... 14 1.7. SDWMN gateway...... 15 1.8. The WMN topology measured using traceroute from a research device (blue circle) toward the other research devices (red circles). The SDWMN controller and the gateways are deployed on the research devices...... 18 1.9. Shared bandwidth utilization...... 19 1.10. Accumulated serving rate...... 19 1.11. Shared bandwidth utilization for diverse flow arrival rates across 250 seconds experimentation time...... 19 1.12. Accumulated serving rate for diverse arrival rates across 250 seconds experimentation time...... 19 1.13. Setup time per flow...... 20 1.14. Control communication overhead...... 20

2.1. LabBrick Digital Attenuator by Vaunix...... 24 2.2. Kernel and User-Space Components of RegMon...... 25 2.3. MAC-state distribution with UBNT Nanostation M5 CONFINE platform at 5GHz channel 40...... 26 2.4. Typical setup with USB connected Laptop ...... 27 2.5. Minstrel-Blues adaptation at different attenuation levels over time ...... 27 2.6. Minstrel-Blues setup for desktop experiments ...... 28 2.7. Throughput and mac80211 states over time with TPC enabled/disabled...... 29 2.8. Transmitpower and SNR over time with TPC enabled/disabled ...... 30 2.9. Minstrel-Blues Setup at UPC ...... 31

3.1. Architecture of the network characterisation daeomon ...... 36 3.2. Discovery of the nodes in a mesh network and the links’ properties in four steps using the NCD user interface (NCui)...... 38 3.3. Screenshot of the NCui showing the mesh graph and highlighting the path from the far right node to the one on the far left...... 39

4.1. The INTERFERENCEHELPER splits frames into constant SNIR chunks and the current NS-3 interference model treats overlapping frames simply as if they were temporary noise contributions...... 44

4 List of Figures List of Figures

4.2. Message-in-message constellations: We consider two equally sized frames with different power levels, separated by a defined signal-to-interference ratio (SIR). The difference of arrival times is specified by the arrival delay of the high-power frame.. 45 4.3. Frame Error Ratio (FER) of the high-power frame for several SIR points observed in a simple two-frame-overlap simulation experiment (cf. Fig. 4.2). This figure also shows that the default NS-3 model does not contain any physical-layer capture model yet...... 46 4.4. Configuration used for measurement-based analysis of the reception performance under interference and the capture effect in particular. Therefore, two individually delayed IEEE802.11 OFDM frames are superimposed in radio frequency domain and fed to the receiver input of the DUT...... 47 4.5. The baseband signal after the passive power combiner for three exemplary delay settings (down-converted to digital baseband by a third USRPN210 used for monitoring purposes). Eventually, the measurement system allows to reproduce arrival time differences with an accuracy of less than 40 nanoseconds...... 48 4.6. Setting a digital intermediate frequency of 10MHz was necessary in order to shift detrimental artifacts (the local oscillator power at DC) out of the designated frequency band...... 48 4.7. Measured frame error ratio of the high-power frame for several SIR points. Compar- ing this figure with Fig. 4.3 shows a systematic difference of 4dB in required SIR. . . 49 4.8. Measured frame error ratio of the high-power frame for higher SIR points. It shows that this particular receiver implementation is able to capture delayed high-power frames as soon as the SIR exceeds around 8dB and reaches capture probability close to one at SIR=12dB...... 50 4.9. Flow chart for the processing of a start frame event in NS-3. The shaded part shows the additional processing for frame capture events. Without capture (default NS-3), a new frame that starts while in receive mode is always dropped...... 51 4.10. This figure shows the channel busy ratio readout at 8 co-located nodes that are operating on channels 3 to 10. At time 40 a nearby AP-client pair starts to transmitter on channel 3. The spillover to adjacent channels (up to channel 8) is clearly visible. . . . 54 4.11. Current versions of Wireshark support parsing of the QBSS beacon element as defined in [1], allowing us to validate our current implementation...... 55 4.12. Current implementation of the beacon add qbss function. The block at line 16-41 needs to be further improved by reducing its complexity and replacing the rcu read lock – we are not sure if it is save to use that method in the given context on all processing platforms...... 56 4.13. Output of a patched version of the network scanning tool called ’horst’. It shows channel load broadcasted from appropriately modified nodes operating on the same channel but located in different rooms of our office building, giving insight into the global radio resource situation and the spatial isolation between nodes...... 57 4.14. Modified Minstrel-HT modulation and coding scheme statistics, those tables now also contain the number of successfully/not-successfully transmitted bytes to each unicast address...... 58 4.15. Illustration of the concept of a resource-utilization-map in transmit direction. The colored areas correspond to the consumption of radio resources that do not lead to a benefit in terms of bits being successfully propagated through the network...... 58

Deliverable D.4.7 5 List of Figures List of Figures

5.1. Active nodes map ...... 63 5.2. Interactions Map ...... 64 5.3. Field notes template ...... 66 5.4. Communicative Ecology of Km2 Poblenou (May 2015) ...... 66 5.5. Sharing projects via social media ...... 69 5.6. Social Media Analytics ...... 69 5.7. Communicative Ecology (Layers) ...... 70 5.8. Communicative Ecology Network Data (May 2015) ...... 70

6.1. All GuifiNet Working Nodes by Year Created ...... 80 6.2. GuifiNet Working Nodes Created 1999-2005 ...... 81 6.3. GuifiNet Working Nodes Created 2006-2008 ...... 82 6.4. GuifiNet Working Nodes Created 2009-2010 ...... 83 6.5. GuifiNet Working Nodes Created 2011-2012 ...... 84 6.6. GuifiNet Working Nodes Created 2013-2014 ...... 85 6.7. GuifiNet Working Nodes With Population ...... 86 6.8. GuifiNet Working Nodes With Population, Barcelona ...... 87

7.1. Distribution of the co-operative expenses...... 102

8.1. Map of deployed Research Devices in the ninux Rome community network (left) and Research Device deployment example (right) ...... 106 8.2. VXLAN L2 Topology ...... 107 8.3...... 110 8.4. Experiment Scenario ...... 112 8.5. The topology of the network used for the experiment, inferred through the execution of the traceroute command on the slivers. Each node in the graph represents a Layer 3 hop. The round intermediate orange nodes represent Community Devices, while the rectangular nodes represent Community-Lab slivers. The ”OF edge”, ”proxy” and ”server” slivers are part of an OpenFlow overlay network, built using OFX...... 113 8.6. Results of the MPEG-DASH client strategy comparison experiment ran in Ninux.org using Community-Lab with the OFX extension ...... 115

9.1. node installation in Tsapournia village ...... 119 9.2. Tsapournia node RD installation ...... 120 9.3. Sarantaporo.gr CONFINE testbed nodes ...... 120 9.4. slivers allocation by node in the CONFINE Testbed ...... 121

Deliverable D.4.7 6 List of Tables

4.1. Measurement Parameters ...... 49

6.1. NDT tests by location ...... 87 6.2. NDT test results by location ...... 88

9.1. Backbone nodes and research devices of Sarantaporo.gr CWN ...... 119

7 Part I.

Technological experiments

8 1. Cosmos - CrOwd-Shared Mesh netwOrk for universal Internet Service

1.1. Introduction

The Internet today is facing the challenge of a growing digital divide, i.e., an increasing disparity between those with and without Internet access [2]. Access problems often stem from populations living in remote or rural locations where it is not cost-effective for the ISPs to provide Internet access, and from the high cost of acquiring Internet access in comparison to the monthly income in deprived communities. The United Nations revealed the global disparity in fixed broadband access, showing that access to fixed broadband in some countries costs almost 40 to 100 times their national average income [3]. To extend Internet access to the economically excluded communities, there have been several initiatives to build community-led crowd-shared wireless networks, in which home broadband owners share a portion of their home broadband with friends, neighbours, or other users either for free or as part of a service offering by the ISP (e.g., [4,5]). Public Access WiFi Service (PAWS) [4] currently under deployment in a deprived community in Nottingham is an example of such crowd-shared WiFi service. PAWS uses a set of techniques that make use of the available unused capacity in home broadband networks and allowing Less-than-Best Effort (LBE) access to these resources. It adopts an approach of community-wide participation, where home broadband subscribers are enabled to donate controlled but free use of their high-speed broadband Internet to fellow citizens. PAWS provides the research community with a wealth of information on the needs of under-privileged users in terms of their access patterns and of what they use Internet access for. PAWS has been facing ongoing deployment challenges, such as limited coverage, stemming from user sharing patterns. In particular, during the PAWS trial deployment, it was observed that home users did not share their broadband connection over periods at which either the whole bandwidth or all the ports of the home router were needed (i.e., PAWS uses an access point connected to the home router for Internet access sharing). Essentially, PAWS is a crowd-shared network with a single point of access per guest, and as such, Internet access sharing is highly dependent on user sharing policies (i.e., the periods at which user share their Internet connection). To mitigate this problem, we investigate the potential benefits of extending PAWS or any crowd- shared wireless network to a wireless mesh network (WMN) by interconnecting wireless home routers. As such, a crowd-shared WMN provides extended coverage via multiple points of access for each guest. We particularly consider crowd-shared WMNs in residential areas, taking advantage of the dense deployment of wireless home routers.

1.2. Background

In this section, we present an SDN architecture to manage crowd-shared wireless mesh networks. In particular, we start by discussing the problem of single-point Internet access sharing on crowd-shared network and subsequently present the beneﬁts of SDN crowd-shared WMN. Next, we provide an

9 1.2. Background 1. Cosmos

no sharing sharing

sharing sharing

Internet

guest user

Figure 1.1: Crowd-shared WMN for public Internet access. overview of the different components of our architecture. Subsequently, we delve into details and propose an algorithm for trafﬁc redirection which takes into account the user sharing policy. Finally, we introduce the implementation of our SDN control plane and our Internet access gateway.

1.2.1. Software Deﬁned Crowd-Shared Wireless Mesh Networks

The underlying problem with PAWS or any crowd-shared network is that they serve as single point of Internet access to guests within the coverage of the wireless router and hence, they have no provision to extend the coverage when no bandwidth is being shared. Based on our experience from the trial PAWS deployment, PAWS routers were not available for certain periods, because sharers needed all the bandwidth of their broadband connection or due to other reasons, such as economic constraints placed on home users in underprivileged areas where they are enforced to conserve energy by turning off the routers at nights. These observed user behaviors entail signiﬁcant challenges for the successful adoption of PAWS. A potential solution to this problem is to extend the PAWS network as a crowd-shared WMN. Such a network would allow home network users to share part of their own broadband connection with the public for free while also connected to each other as a WMN providing extended coverage (Fig. 1.1). Extending PAWS to a crowd-shared WMN departs from the norm: multiple users from different ISPs form part of the WMN to provide free Internet connectivity, while most wireless community WMNs today are operated by a single organization. This raises important questions regarding the operation, conﬁguration, and management of crowd-shared WMNs. SDN can facilitate the management and operation of wireless networks at large scale. Leveraging on SDN’s centralized control and network-wide visibility, the management and operation of a crowd- shared WMN can be outsourced to a third party. In [6], we describe a holistic approach of coupling both social and economic incentives in designing future networks, allowing the extension of the stakeholder value chain to include more than the two traditional parties (consumer and Internet Ser-

Deliverable D.4.7 10 1. Cosmos 1.2. Background vice Provider). Compared to existing mesh networks for Internet access sharing (e.g., Freifunk [5]), our approach provides more opportunities for non-governmental organizations and local governments (driven by social goals rather than economic) to become virtual network operators. Enabling a third party to federate such wireless home networks would reduce the operating expenditures for network operators as well as enable new economic models for revenue generation from currently underutilized infrastructures. In particular, we rely on SDN to create the notion of Virtual Public Networks (VPuN), i.e., crowd- shared home networks created, deployed and managed through an evolutionary SDN control abstraction [6]. Although originally intended for crowd-shared wireless networks such as PAWS, VPuN can be also used for crowd-shared WMNs, enabling resource pooling across multiple home broadband connections based on the prevailing network conditions and usage sharing patterns. Based on the notion of VPuN, we have deployed a SDWMN control plane in Athens Wireless Metropolitan Network (AWMN), i.e., one of the community networks of the CONFINE project. AWMN contains research devices (RDs) that host isolated containers (i.e., so-called slivers) which can be allocated and controlled by a user, according to the needs of his experiment [7]. It has a WMN with approximately 30 RDs and more than 1000 wireless nodes (each RD is connected to the WMN). Fig. 1.2 illustrates an overview of the SDWMN control plane. The main goal of the control plane is to improve the utilization of shared bandwidth by redirecting guest traffic to one of the home routers through the WMN, based on user sharing policies. As such, the control plane exposes an interface to each home user, allowing him to express a sharing policy (e.g., using the sharing policy expression language presented in [6]). The control plane functionality mainly consists of traffic redirection and monitoring. The traffic redirection module selects the gateway for Internet access (using the algorithm presented in Section 1.2.2) and subsequently creates a tunnel between the home router where the guest is connected and the router assigned as the gateway (home routers are deployed in the CONFINE RDs). Traffic is redirected through a tunnel, since the wireless nodes in the AWMN community networks are not under our control (as opposed to the RDs). A tunnel is dynamically set up when traffic has to be redirected over that path, and it is torn down when it is no longer needed. The monitoring module collects statistics about the bandwidth utilization of the WMN and the access links. A detailed description of the implementation of the SDNWM control plane and the gateway is given in Section 1.2.4. In AWMN, we deploy the control plane in one of the RDs. However, in real world depolyment, we envision hosting the SDN controller as a virtual machine in micro-datacenters hosted by large ISPs [8]. Micro-data centers are small DCs deployed at POP and network aggregation points. By deploying the controller in micro-DCs we ensure low communication delay with the home routers and scalable processing power to cope with high loads (scale up the controller computation capacity by using more servers).

1.2.2. Trafﬁc Redirection

Assigning flows to network paths with constrained capacity can be formulated as a multi-commodity flow problem which is known to be NP-hard [9]. Hereby, we present a heuristic algorithm for the assignment of the gateway and the path over which the traffic will be redirected through the WMN. The algorithm aims at maximizing Internet shared bandwidth utilization by accommodating an as large as possible number of flows. It is executed by a SDN controller which has knowledge of the WMN topology and utilization (see section 1.2.4), the Internet access link utilizations, and the user sharing policies.

Deliverable D.4.7 11 1.2. Background 1. Cosmos

Sharer Virtual Network Operator

Northbound API

Traffic Redirection Monitoring

SDN Control Plane Southbound API (OpenFlow, SNMP)

Research Device Research Research Device Device

Community Network

Research Research Device Device

Figure 1.2: Software-deﬁned WMN control plane overview.

The traffic redirection algorithm assigns a gateway and the shortest path to the arriving flows. Each flow has a rate and a local home gateway. The algorithm is executed whenever there is insufficient shared bandwidth in the local home network. Initially, the flows are sorted in decreasing order based on their rates. For each flow, the algorithm selects the home router with the highest access link bandwidth. Subsequently, the algorithm identifies the set of paths between the local home router and the assigned gateway that satisfy the flow rate. In case there is no such path, the algorithm performs another iteration for the selection of the gateway, excluding the previously selected home router. Otherwise, the shortest among these paths is being identified based on the number of hops. This eventually computes the path for the traffic redirection through the WMN.

1.2.3. User Sharing Policies

The traffic redirection algorithm assigns flows to gateways taking into account the amount of shared bandwidth and optionally the path hop-count. So far, we have not exploited any advance knowledge of user sharing policies for traffic redirection. Such information can prevent wasteful traffic redirections (i.e., forwarding traffic to a router which will shortly become unavailable for Internet access sharing). Sharing policies can be specified using a sharing policy expression language (SPEL), e.g., the XML- based schema presented in our previous work [6]. SPEL essentially provides the necessary means to the sharers to specify the amount of shared bandwidth as well as the period of sharing. This information is conveyed to the SDN control plane and is used as an additional input for gateway assignment. We run a trace-driven simulation of a crowd-shared WMN to investigate the impact of sharing policies on traffic redirections. In particular, we model the availability of the home routers using an on-off

Deliverable D.4.7 12 1. Cosmos 1.2. Background

1600

1400

1200

1000

800

600

number of re-assigned flows 400 without GW availability knowledge with GW availability knowledge 200 5 10 15 20 25 30 35 40 45 50 flow arrival rate (flows/min) Figure 1.3: Flow redirections.

Markov chain. On and off times are exponentially distributed with mean values µon = 106 minutes and µoff = 555 minutes. We parameterize the exponential distributions using datasets from the PAWS deployment in UK. Fig. 1.3 illustrates the number of traffic redirections for different network loads with and without the knowledge of sharing policies. When sharing policies are known, the traffic is directed to the router with the longest period of availability (among the routers with sufficient shared bandwidth). According to Fig. 1.3, using sharing policies leads to fewer traffic redirections, especially for low and moderate network loads. For higher loads, the number of routers with sufficient shared bandwidth diminishes, and as such, the knowledge of sharing policies does not yield significant gains. We perform additional tests to investigate potential implications, such as packet reordering and higher communication overhead, due to traffic redirections. First, we use the experimental setup depicted in Fig. 1.4 to quantify the level of packet reordering. In this setup, UDP traffic is redirected from path 1 to path 2 at the intermediate node. The redirection is carried out by inserting flow entries to an OpenvSwitch (OVS) datapath that has been set up in the intermediate node. We further employ a delay node, set up in path 2, to adjust the delay along this path. Fig. 1.5 shows that the number of out-of-order packets increases in proportion to the delay difference between the two paths. Our path delay measurements in the CONFINE community network indicate that a redirection through the WMN increases the RTT with up to 50 ms. This will result in the out-of-order delivery of a small number of packets, which, in turn, can lead to congestion control invocations, and, hence, lower throughput for TCP flows. Besides packet reordering, traffic redirections generate additional communication overhead between

Deliverable D.4.7 13 1.2. Background 1. Cosmos

225 1 Mbps 200 2 Mbps 5 Mbps 175

150 path 1 125 100 Src OVS Sink 75

50 number of packets out−of−order path 2 25

0 0 100 200 300 400 500 delay node (tc) delay(old_path) − delay(new_path) (ms) Figure 1.4: Experimental setup for packet Figure 1.5: Packet reordering. reordering evaluation.

repository traffic redirection gateway selection

monitoring

flow table tunnel BW path gateway configuration installation measurement measurement registeration

Southbound API (OpenFlow, XML-RPC)

To gateways

Figure 1.6: SDWMN controller. the SDWMN controller and the routers. Specifically, for each flow redirection the router sends a OFPT PACKET IN message to the controller, which, in turn, sends an OFPT FLOW MOD message to the router. Taking the respective acknowledgment messages into account, the communication overhead per flow redirection is 490 bytes. This overhead can increase substantially with a large number of guest flows per home router, as the PAWS router logs indicate. Considering the identified implications of traffic redirections, we employ sharing policies for the assignment of guest flows to gateways.

1.2.4. Implementation

We provide a detailed description of the implementation of the SDWMN control plane and gateway in Sections 1.2.4.1 and 1.2.4.2, respectively.

1.2.4.1. SDWMN controller

We implement our controller in Python using POX which provides a framework to implement SDN controllers with OpenFlow protocol interface [10]. Since OpenFlow supports only the conﬁgura-

Deliverable D.4.7 14 1. Cosmos 1.2. Background

control gateway controller data

repository incoming redirected traffic Locally served traffic outgoing redirected traffic

encapsulation (Click) Internet port

local guest traffic

encapsulation (Click) XML-RPC server XML-RPC encapsulation (Click)

path measurement (ping, traceroute)

decapsulation (Click)

encapsulation (Click)

virtual ports

switching (Openvswitch)

physical port

Figure 1.7: SDWMN gateway. tion of switch flow tables, we use XML-RPC interface to install and configure tunnels between the gateways. Our SDWMN controller consists of the following modules (Fig. 1.6): • Gateway registration: Each new gateway joining the crowd-shared network is registered with the creation of a new gateway object in the controller repository. Each gateway object holds information about the gateway’s: i) datapath ID, which is a 64 bit unique identifier for each OpenFlow switch instance, ii) IP address of the WMN port, iii) ports table, which contains the names and OpenFlow switch ports numbers (i.e., DSL port, the WMN port and the guests’ wireless network port), iv) monitoring table, which contains the gateway shared bandwidth utilization and WMN path quality, and v) tunnel table, which contains the IP addresses, UDP port numbers (we use UDP-in-IP tunnelling) and OpenFlow switch port numbers of the installed tunnels. The registration process is triggered by POX when the gateway establishes a new control channel with the controller. • Monitoring: This module collects and processes the measurements of shared bandwidth utilization and the WMN paths delay and hop counts for each gateway. Bandwidth (BW) estimation in a WMN is very challenging, due to BW fluctuations and the low degree of accuracy of available BW measurement tools (e.g., pathload [11], TOPP [12], and IGI/PTR [13]) that was also confirmed by our own tests. However, in our experimental setup, the Internet access links are the bottleneck (since they have been configured at lower capacity than the WMN), and, as such, the aforementioned WMN bandwidth estimation issues do not affect the accuracy of our measurements. Using the OpenFlow protocol, the BW measurement module pulls the network port counters of each home gateway to acquire the accumulated number of bytes received/sent in the network port (e.g., the DSL port). In particular, the controller sends an OpenFlow OFPT STATS REQUEST message with OFPST PORT type and the port number to the gateway. The gateway replies with OFPT STATS REPLY message which carries the latest reads of the gateway rx bytes and tx bytes counters. To calculate the shared bandwidth utilization, the BW measurement module applies exponential moving average to the counters changes

Deliverable D.4.7 15 1.2. Background 1. Cosmos

over time. In addition, the path measurement module uses XMP-RPC to pull the RTT and hop- count measurements of each WMN path to/from each gateway. All measurements are stored in the monitoring table of the respective gateway object. • Traffic redirection: Flows are redirected using the following modules: – Gateway selection: The gateway selection module is invoked when a gateway sends a OFPT PACKET IN message to the controller to announce the arrival of a new flow. Based on the measurements (shared BW utilization and path length) provided by the monitoring module and using the algorithm described in Section 1.2.2, the gateway selection module assigns a gateway to the new flow. Subsequently, a tunnel is being set up (see tunnel installation for further details) between the home gateway (where the flow arrives) and the assigned gateway (where the flow is redirected). Next, the flow table configuration module installs the respective forwarding entries in the home and the assigned gateway such that the guest flow is redirected accordingly. – Tunnel installation: Using the XML-RPC interface, it installs tunnel modules (encapsulation module) in a gateway. In particular, it sends the IP address of the assigned gateway to the flow home gateway as well as the source and destination port numbers of the UDP header used for encapsulation. Upon successful installation of the tunnel module, the home gateway replies with the switch port number of the new encapsulation module. This information is stored in the tunnel table of the home gateway object to be used later by the flow table configuration module. – Flow table configuration: This module installs the corresponding flow entries in the home gateway and the assigned gateway using OpenFlow. This consists in sending a OFPT FLOW MOD message to the home and assigned gateways. This message carries the flow matching fields [10](e.g., source/destination IP, port source/destination) as well as the OpenFlow switch output port number.

1.2.4.2. SDWMN Gateway

Our SDN gateway exposes to the SDWMN controller an OpenFlow and XML-RPC interface to install tunnels and redirect flows (see Fig.1.7). Tunnels are created by encapsulating guest traffic in IP-in- UDP packets using the encapsulation module (red line Fig. 1.7). The destination IP address of the tunnel header is set to the IP address of the assigned gateway. For each new tunnel a new encapsulation module is created. On the other hand, incoming tunnelled packets (green line Fig.1.7) are processed by the decapsulation module to strip their IP-in-UDP header before they are delivered to the Internet port. Both encapsulation and decapsulation modules are implemented using Click [14]. To steer data traffic between the physical ports and the modules (encapsulation/decapsulation), we rely on OpenvSwitch [15]. This switching module also forwards messages from the SDN controller to the gateway controller module through the XML-RPC server. To connect modules to OpenvSwitch, we use Linux TAP devices. The instantiation of encapsulation/decapsulation modules is performed by the gateway controller in the following steps: (i) creation of TAP interfaces, (ii) appending TAP interfaces names, MAC and IP address of the assigned gateway to template Click configuration files stored on the repository (we use templates to speed up modules instantiation) and (iii) installation of the Click configuration. Furthermore, the gateway controller collects and transfers the measurements generated by the path measurement module to the SDN controller. The path measurement module uses ping and traceroute

Deliverable D.4.7 16 1. Cosmos 1.3. Experiment Description to collect RTT and hop counts of the WMN paths. A list of gateways IP addresses is passed to this module to obtain the measurements. The module sends 10 probing packets every 30 seconds to avoid ﬂooding the network.

1.3. Experiment Description

To quantify the benefits of a crowd-shared WMN (compared to crowd-shared network with a single point of access, such as PAWS), we have deployed and configured a SDWMN in the Athens Wireless Metropolitan Network (AWMN). In our experimental study, we have measured the following: • Shared bandwidth utilization: We aim at measuring the total shared bandwidth utilization of all access gateway of crowd-shared network with WMN and without WMN. This experiment shows the improvement in crowd-shared network resource utilization (the shared BW utilization) when extended with a WMN. • Accumulated serving rate: For a crowd-shared network with and without WMN, we quantify the accumulated serving rate ASR(T ) at time T , defined as:

ΣT sfinished ASR(T ) = t t , (1.1) T finished T rejected (Σt st + Σt st )

finished where st denotes the total sizes of the flows at time t which are accepted (i.e., assigned rejected Internet access) and successfully served without disruption due to router unavailability. st denotes the total sizes of the flows at time t which are rejected (i.e., not assigned Internet access). The objective of this experiment is to depict the increase of the total accumulated crowd-shared capacity when a WMN is introduced. • Control overhead: To study the scalability of our control plane, we measure the total control overhead in terms of BW consumption for diverse flow arriving rates. We further break down the control BW into monitoring and flow setup delay. • Flow setup delay: Latency is a critical aspect for several applications (e.g. VOIP, video). Therefore, we evaluate the additional delay incurred by our controller to setup flows when they arrive to the crowd-shared network. We run this experiment for different flow arriving rates to illustrate the scalability of our control plane.

1.4. Test Setup and Results

For our experiments, we use 12 research devices (RDs) to deploy 11 home routers (or gateways) and one SDWMN controller. Fig. 1.8 illustrates the WMN topology measured from a single research device. We implemented a tool to analyse and construct the topology out of traceroute measurements. Using MGEN [16], we generate guest ﬂows with a rate and lifetime sampled out of a uniform and an exponential distribution, respectively. The ﬂows arrive to the network according to a Poisson process. For router availability pattern, we use the model described in Section 1.2.2. Since running tests for more than 555 minutes (the mean value of the off period) is not feasible, we scale down the mean values to µon = 10.6 seconds and µoff = 55.5 seconds. In our experiments, we emulate the router status by sending a signal to the controller which indicates the availability or unavailability of a router. The time required to generate the signal is negligible (a few milliseconds). Due to occasional reliability issues in the RDs, we restricted the duration of our experiments to 250 seconds. After

Deliverable D.4.7 17

1.4. Test Setup and Results 1. Cosmos *

Research device

Research device running traceroute

Intermediate hops (wireless routers) * paths with no ICMP responce Figure 1.8: The WMN topology measured using traceroute from a research device (blue circle) toward the other research devices (red circles). The SDWMN controller and the gateways are deployed on the research devices. downscaling the router on/off periods, there are several transitions between the router states during each experiment. We initially measure the shared bandwidth utilization (Fig. 1.9) and the serving rate (Fig. 1.10) with an arrival rate of 60 flows per minute, across 10 runs. Variations in the measurements across the different runs were insignificant. Fig. 1.9 illustrates a low utilization of the shared bandwidth without a WMN during the whole period, although there is high demand for Internet access by guests attached to the various home networks. In contrast, a WMN allows to capitalize the unused capacity and accommodate a larger volume of guest traffic. More precisely, according to Fig. 1.9 guest traffic redirection through the WMN results in the full utilization of the shared bandwidth. Furthermore, crowd-shared WMNs can accommodate substantially larger volume of guest traffic, as depicted in Fig. 1.10. This stems from the high utilization of the shared bandwidth. We further measure the shared bandwidth utilization and serving rate with a wide range of guest traffic demands. In this respect, the boxplots in Figs. 1.11 and 1.12 illustrate the shared bandwidth utilization and serving rate with diverse flow arrival rates, ranging from 20 to 80 flows per minute1. These results corroborate the efficiency of the WMN for various traffic loads, as the shared bandwidth utilization and the serving rate always remain very high. In Fig. 1.12, the flow arrival rates of 60 flows/min and beyond lead to request rejections, due to the insufficient access link bandwidth. On the other

1 all boxplots show measurements across the time period of the experiment, i.e., each point on the graph represents a different time point. For different runs we calculate the average, the variation is insigniﬁcant.

Deliverable D.4.7 18 1. Cosmos 1.4. Test Setup and Results

1.2 1.0 with WMN with WMN without WMN without WMN 1.0 0.8

0.8

0.6

0.4 0.4 shared BW utilization accumulated serving rate 0.2 0.2

0.0 0.0 0 50 100 150 200 250 0 50 100 150 200 250 time (sec) time (sec) Figure 1.9: Shared bandwidth utilization. Figure 1.10: Accumulated serving rate.

1.2 1.0 with WMN without WMN 0.9 1.0

0.8

0.8 0.7

0.6 0.6

0.5 0.4

shared BW utilization 0.4 accumulated serving rate 0.2 0.3 with WMN without WMN 0.0 0.2 20 40 60 80 20 40 60 80 flows arrival rate (flows/min) flows arrival (flows/min) Figure 1.11: Shared bandwidth utilization for Figure 1.12: Accumulated serving rate for diverse ﬂow arrival rates across 250 seconds diverse arrival rates across 250 seconds experimentation time. experimentation time.

hand, Figs. 1.11 and 1.12 show poor bandwidth utilization and serving rate without a WMN. In this particular case, the restriction of one point of Internet access for each guest leads to shared bandwidth wasting. Essentially, our results show the significant benefit that a WMN can bring into crowd-shared networks, by effectively pooling resources across all home networks. To evaluate the scalability of our control plane, we measure the flow setup time and the average control communication overhead across a range of flow arrival rates. We define the setup time as the interval between the flow’s first packet arrival at the gateway and its transmission to the next hop. This is essentially the time incurred for gateway selection and flow entry installation. We also define the control communication overhead as the bandwidth consumed to setup flows and monitor the shared bandwidth utilization. We run our experiments on a single gateway (the results can be

Deliverable D.4.7 19 1.5. Conclusions 1. Cosmos easily extrapolated for different number of gateways) and generate 100 different flows for each arrival rate. Figs. 1.13 and 1.14 show that per flow setup time does not change significantly, while the control communication overhead increases linearly and is in the range of tenths of Kbps. According to these results, the performance of our controller scales with increasing control loads.

60 40 flows setup traffic 35 monitoring traffic 50

30 40 25

30 20

15 20

10 5 per flow setup time (milliseconds)

0 control communication overhead (kbit/sec) 0 60 180 300 420 540 60 180 300 420 540 flow arrival rate (flows/min) flow arrival rate (flows/min) Figure 1.13: Setup time per ﬂow. Figure 1.14: Control communication overhead.

To quantify the benefits of crowd-shared WMN on a larger scale, we developed a simulator (in Python) that models the flow-level behavior of the guest traffic in a crowd-shared WMN. Using simulation, we again compared the shared bandwidth utilization and the serving rate with and without a WMN for different size of topologies. Our simulation results confirmed our experimentation findings. In particular, for different size of topologies, the crowd-shared network with WMN showed high shared BW utilization and high serving rate in comparison to the crowd-shared network without WMN.

1.5. Conclusions

In COSMOS, we investigated the benefits of extending the coverage of any crowd-shared network (e.g., PAWS) by connecting the home routers as a mesh. A crowd-shared WMN can mitigate the fundamental problem of any crowd-shared network, i.e., the presence of a single point of access for each guest. In a crowd-shared WMN, the path redundancy towards the Internet backhaul can be exploited to achieve better resource utilization and open Internet access for larger number of users, especially during periods of limited Internet access sharing. We showed that the advance knowledge of user sharing policies can reduce the number of flow redirections (particularly for low and moderate network loads) avoiding implications, such as packet reordering and communication overhead. By using an SDN-based controller to incorporate the sharing policies in a simple centralized algorithm, we are able to perform traffic redirection leading to a high utilization of the shared bandwidth, as opposed to a crowd-shared network with a single point of access where a significant amount of shared bandwidth is wasted. Eventually, an SDN crowd-shared WMN can accommodate a substantially larger volume of guest traffic. We further investigated the scalability of our SDWMN control plane. In this respect, we showed that our control plane can pro-

Deliverable D.4.7 20 1. Cosmos 1.5. Conclusions cess high number of flows while maintaining low setup times per flow and low control communication overhead. To this end, an SDN crowd-shared WMN can capitalize the residual bandwidth in home broadband connections and consequently, create opportunities for generating revenue from currently underutilized infrastructures. Furthermore, SDN simplifies the process of deploying new algorithms and policies for Internet access sharing by enabling centralized software-based upgrade. As our focus is on maximizing shared BW utilization, other algorithms aiming, for instance, to preserve users’ privacy by introducing redirection geographical (neighbourhood or building) or social (only friends) constraints might also be deployed (e.g., algorithms redirecting through local services, such as caches). We believe that SDN opens up opportunities to provide better services in crowd-shared networks. This can incentivize ISPs to participate and further promote such Internet access sharing models.

Deliverable D.4.7 21 2. MinstrelBlues - Joint Transmit Power and Rate Control on CONFINE Testbed

2.1. Introduction

IEEE 802.11 (WiFi) networks are used to provide Internet access anywhere anytime. However, their performance is far below the achievable limits when multiple participants share the same frequency spectrum in an uncoordinated manner. The major reason behind such inefficiency is the lack of practical resource allocation algorithms that adapt well to wireless network dynamically and select the appropriate transmission parameters such as transmission rates and power levels. Most current practical schemes are rather simplistic and only change a single transmission parameter. For instance, Transmit Power Control (TPC) works at the WiFi PHY layer and commonly assigns a static and rather high power level to all packets regardless if a lower transmission power would provide an equal network performance with lower interference to its environment. Someone can compare general WiFi communication between access points and clients with a group of people having a pair- wise conversation on a big round table. Today’s state of the art is that all people shout at a fixed level of loudness to their communication partner and potentially interfere other conversations. With our proposed transmit power control we can significantly reduce the needed volume (in WiFi terms: transmission power) per communication link while providing the same data rate. Now people at both ends of the table can have a their conversation in parallel, as they do not interfere anymore. A per-link or packet power control scheme is expected to provide better network performance through less interference, but typically increases complexity and requires higher-layer information, such as medium access state from the Medium Access Control (MAC) layer. Therefore, although performance im- provements have been shown in theory, these ideas are largely uninvestigated in practice. Our main goal is to extend our joint rate and power controller Minstel-Blues from current IEEE 802.11 a/b/g support to actual IEEE 802.11 n/ac network devices within the mac80211 subsystem of the Linux kernel. After we have validated our implementation with a digital attenuator desktop like setup, we want to measure its performance in different scenarios - from 1-link desktop up to 2-links within the CONFINE UPC environment. In order to achieve efficient rate and power control in IEEE 802.11 wireless networks, an important requirement is to understand their impact on channel utilization and network throughput. Improper power, rate, and carrier sensing adjustments can lead to under-utilizing or attempting to over-utilize the medium, which can stall communication, or in the worst case result in network breakdowns. As packet level traces are insufficient to directly monitor medium utilization at the MAC-layer, we created a new wireless monitoring tool “RegMon” within the Linux kernel.

2.2. Background

The potential and challenges of transmission power control in wireless networks have been widely discussed in literature. As early as 1998, Bambos states [17]: Transmitter power control can be used to concurrently achieve several key objectives in wireless networking, including minimizing power consumption and prolonging the bat-

22 2. MinstrelBlues 2.2. Background

tery life of mobile nodes, mitigating interference and increasing the network capacity, and maintaining the required link QoS by adapting to node-movements, ﬂuctuating interferences, channel impairments, and so on. Moreover, power control can be used as a vehicle for implementing on-line several basic network operations, including admission control, channel selection and switching, and handoff control.

Today, we can roughly categorize wireless networking: (1) Infrastructure-based networks such as cellular networks or WiFi infrastructure networks or (2) ad hoc networks in WiFi. Note also that, while most infrastructure-based networks assume single-hop communication, in ad hoc networks multi-hop communication is typical. If we view wireless networks as a collection of interfering links, infrastructure-based wireless networks is a simpler case of the more general ad-hoc networks [17]. The power control problem in the general model is especially complex, since the choice of the power level fundamentally affects many aspects of the network operation [18]:

• Received signal quality at the receiver (PHY layer)

• Carrier sensing and data transmission ranges, hence, the number of nodes contending for channel access (MAC layer)

• Transmission range, and hence, the number of hops in routing (Network layer).

• Interference with other receivers, therefore, congestion (Transport layer).

Note that in infrastructure WiFi networks, TPC does not affect the routing layer. Furthermore, in cellular networks, the impact of TPC on the MAC layer is less of an issue due to the use of FDD (Frequency Division Duplex) or TDD (Time Division Duplex) schemes for medium access. The capacity gains of distributed power control in the context of cellular networks have been well studied in the literature [19]. As WiFi uses CSMA/CA MAC these solutions for cellular networks cannot be directly applied to infrastructure-based or ad hoc WiFi networks,

Within our latest research work, we firstly enabled cross-layer communication of transmission power between the WiFi PHY and MAC layers in the Linux mac80211 subsystem. Based on our IEEE 802.11 a/b/g Atheros WiFi hardware we designed and implemented a distributed rate and power control algorithm, Minstrel-Blues, which does not rely on signal strength or channel state information, but uses local statistics from periodic sampling of different rate and power combinations. Essentially, Minstrel-Blues can run on any IEEE 802.11 a/b/g WiFi hardware that supports packet-level power and rate control capabilities. Our results show that if the goal is on maximizing the per-link throughput, Minstrel-Blues can significantly reduce the transmission power necessary to communicate per link, while maintaining the same throughput achieved with maximum transmit power. Our distributed joint rate-power controller is able to increase efficient usage of the shared WiFi spectrum through less interference and increased spatial reuse. In order to leverage Minstrel-Blues network performance gains with actual IEEE 802.11n hardware we need to extend our current implementation as those new WiFi chips use different devices drivers and a separate rate control called Minstrel-HT. As most IEEE 802.11 n hardware support per packet power control we are sure that our proposed Minstrel- Blues algorithms outperforms actual WiFi systems that rely on a static transmit power level for all packets. Due to the limited scale of our experimental network, we developed a simulator (in Python) that models the flow-level behavior of the guest traffic in a crowd-shared WMN.

Due to the limited scale of our experimental network, we developed a simulator (in Python) that models the ﬂow-level behavior of the guest trafﬁc in a crowd-shared WMN.

Deliverable D.4.7 23 2.3. Experiment Description 2. MinstrelBlues

2.3. Experiment Description

To the best of our knowledge, we are the ﬁrst to implement a cross-layer joint power and rate controller Minstrel-Blues, which enables realizing different power control approaches using todays WiFi hardware. Minstrel-Blues has different tuneable parameters that specify the way power and rate is adjusted. The experiment data set from our proposed CONFINE experiments is used to identify reasonable parameter values that work under real WiFi conditions. The collected data provides us a training set for designing an adaptive conﬁguration mechanism that is based on the environmen- tal characteristics. This research investigation allows us to adapt Minstrel-Blues to be incrementally deployable into today’s real world WiFi networks.

RegMon MAC-layer monitoring: For the experiments itself we ﬁrst developed a Minstrel-Blues and RegMon enabled OpenWrt image for our CONFINE research nodes (which are Ubiquity Nano- Stations M5 with Atheros IEEE802.11 n 5GHz WiFi chips). RegMon enables sampling, and analysis of hardware registers of the WiFi MAC-layer with a high accuracy through a sampling rate up to 20 kHz depending on the measurement hardware. It is implemented and validated for our CONFINE research nodes within the ath9k driver. Our measurement tool enables wireless researchers to access new information sources. RegMon essentially provides us with MAC layer measurements that help develop, troubleshoot and analyze our joint rate and power controller. Our experiments with the CONFINE research nodes show that our implementation of RegMon is able to perform ﬁne-grained MAC-Layer measurements to monitor and analyse the WiFi channel conditions.

TPC validation with digital attenuator: To validate our implementation of Minstrel-Blues in a desktop setup (see ﬁgure 2.4) we use a LabBrick digital attenuator from Vaunix, shown in ﬁgure 2.1. This allows us to change the attenuation between two connected nodes in a controlled manner. Hence we are able to validate our transmit power control implementation in terms of proper transmit power level changes.

Figure 2.1: LabBrick Digital Attenuator by Vaunix.

Minstrel-Blues performance experiments: In order to analyze the performance of our joint rate and power controller Minstrel-Blues we focus on the following two experiments: (1) we use our CONFINE nodes in a 1-link scenario on our side at Technical University of Berlin to analyse the performance impact of Minstrel-Blues on the actual link throughput and (2) we experiment with a 2- link setup at the CONFINE testbed at UPC in Barcelona to analyze the performance impact between interfering links.

Deliverable D.4.7 24 2. MinstrelBlues 2.4. Test Setup and Results

2.4. Test Setup and Results

RegMon Monitoring Experiments In order to analyse the MAC-layer state distributions with RegMon we have performed several experiments on our CONFINE research nodes - Ubiquity Nano- Station M5. To perform the monitoring, we leverage the lightweight kernel-to-userspace debug file system (debugfs). This serves two purposes: • it enables a a simple file-based configuration of RegMon for its in-kernel operations from the userspace • the actual trace-file from the kernel can be accessed via standard file read (e.g., with tail, cat) from the userspace. For the actual measurements, it is possible to access Atheros control and status registers stored in the card memory through the PCI bus, as shown at the bottom of the RegMon figure 2.2. Due to the limited scale of our experimental network, we developed a simulator (in Python) that models the flow-level behavior of the guest traffic in a crowd-shared WMN. RegMon is implemented as a single kernel driver patch for each of the supported three drivers, without the need for additional modules, daemons or user-space applications. The trace-file generated by RegMon is currently formatted as space separated list of measurement values, which turned out to be a sufficient format to parse and process the trace file for further analysis.

RegMon specify adjust cyclic read register addresses sampling rate reg_trace ﬁle user

time reg1 ... reg9 reg0 = 0x9321 101 0x53 ... 0x79 space 103 0x11 ... 0x44 ...... 105 0x61 ... 0x51

(2) user-space reg9 = 0x8321 reg_interval = 10 ... debugfs debug ﬁlesystem per WiFi card sampling

kernel TCP/IP Stack packet traces space

non-mac802.11 mac80211 FullMAC driver

SoftMAC Linux drivers madwifi (1) in-kernel ath9k ath5k register sampling RegMon RegMon RegMon

REG_READ( ) ath5k_hw_reg_read( ) OS_REG_READ( ) PCI Bus

WiFi interface Atheros 802.11 n Atheros 802.11 b/g/a WiFi chips WiFi chips

Figure 2.2: Kernel and User-Space Components of RegMon.

From our performance analysis of RegMon on our CONFINE platform UBNT Nanostation M5 using the current ath9k driver we conclude that the maximum sampling rate that produces meaningful results is about 12kHz while the overall system load is at 30%. At 12kHz sampling rate the bandwidth requirements to collect RegMon traces are about 21MBits in that case of uncompressed trace ﬁle transmission and it drops to 5MBits when LZO compression is used. Based on the statistics collected with RegMon, we are able to calculate the percentage of time spent in each MAC state during a given observation duration. The WiFi MAC can change between the following four MAC states over time: (1) tx-busy indicating packet transmission, (2) rx-busy indicating

Deliverable D.4.7 25 2.4. Test Setup and Results 2. MinstrelBlues packet reception, (3) idle indicating the MAC layer is idling and (4) other indicating interference which can not be decoded as packets. In ﬁgure 2.3 we have visualized a typical MAC-state distribution collected with RegMon running within the ath9k driver of OpenWrt on our CONFINE nodes (Ubiquity Nanostation M5). RegMon comes in handy when we want to look at spacial reuse potential be adapting transmission power levels via Minstrel-Blues.

Figure 2.3: MAC-state distribution with UBNT Nanostation M5 CONFINE platform at 5GHz channel 40.

Experimental validation with digital Attenuator As we want to validate our TPC implementation in terms of correct power level changes, we leverage the Vaunix digital attenuator. The attenuator enables us to simulate a drop or increase in the quality of the connection by changing the attenuation between two wireless devices connected with it. For our test setup we use two Ubiquiti Nanostation M5 with OpenWRT installed. One is conﬁgured as access point, the other Nanostation acts as a station. Their WiFi antenna pins are connected with the attenuator, so they have to communicate through the Lab Brick device. As we only have one attenuator available, only one antenna pin out of the dual stream Ubiquiti device is connected to the attenuator. Both Nanostations are connected by their Ethernet port to a different subnet as the WiFi cards to make changes and take measurements without interfering with our experiment. In addition to the two Nanostations M5 our server for measurements is connected to the attenuator via USB to interfere with the attenuation of the connection.

Deliverable D.4.7 26 2. MinstrelBlues 2.4. Test Setup and Results

Figure 2.4: Typical setup with USB connected Laptop

With our chosen setup we were able to validate our Minstrel-Blues implementation in terms of its ability to adapt to different channel conditions. As shown in ﬁgure 2.5 while we have changed the attenuation from 0dB to 40 dB every 20 seconds, Minstrel-Blues reacts to the varying channel conditions accordingly. The increase and decrease in throughput of Minstrel-Blues is comparable to plain Minstrel and shows the same performance.

Figure 2.5: Minstrel-Blues adaptation at different attenuation levels over time

To test our Minstrel-Blues implementation on current IEEE 802.11n devices we conduct several one and two link experiments with our CONFINE nodes. Those four Ubiquity Nanostations M5 are ﬂashed with a Linux OpenWrt image that includes our RegMon and Minstrel-Blues patches for the ath9k driver. With those four nodes we are able to set-up two wireless links, all nodes are running in adhoc mode, as this is the typical wireless mode for mesh network nodes.

1-link Minstrel-Blues Experiments The 1-link setup is performed indoor in our DAI laboratory at Technical University of Berlin. Both Nanostations are connected to a Ubuntu measurement server by Ethernet. This setup enables us to generate TPC and UDP traffic via Iperf on the measurement machine and reduce the load on the wireless nodes to pure WiFi traffic transmission. Figure 2.6 shows our desktop setup, where both nodes are about 20m apart from each other. We used the indoor channel 40 for our experiments, which is equivalent to 5200MHz. By using Iperf 2.05 on the multi-homed measurement machine we generated two TCP flows that go through the wireless link for 600 seconds. Our Minstrel-Blues implementation allows us to switch the TPC part of our algorithm on and off. This

Deliverable D.4.7 27 2.4. Test Setup and Results 2. MinstrelBlues allows us to compare the throughput in both cases, with and without power control. The experiment starts with TPC disabled for 150s, between the 150th second and 300s there is TPC enabled, after 300s we disables TPC and enabled it again after 450s until 600s. In order to analyse the performance of our alogorithm we captured PCAP packet traces from a monitor interfaces on each Nanostation. And also the RegMon traces are collected from all wireless nodes. All traces are transfered to the measurement machine via Ethernet, as the nodes itself do not have enough disk storage.

Figure 2.6: Minstrel-Blues setup for desktop experiments

After several experimental trials we have merged, parsed and analyzed the dataset. The perfromance results are presented in four plots. Figure 2.7 shows two plots, the upper one summurizes the throughput over time for both TCP flows and the lower plot shows the MAC-state distribution over time. As we can see from the box plots in the throughput graph, the sum throughput of both TPC flows stays constant over the entire 600s experiment duration and the MAC-states are equaly distributed. By comparing the power levels used to transmit the data in figure 2.8 we can see that in both phases where TPC is enabled, Minstrel-Blues does reduce its power level significantly. The lower plot in figure 2.8 show the sender perspective on the used power level for transmission, while the upper plot show the received SNR level of packets at the receiver. The sender is able to reduce its transmission power by about 10dB without a reduction in link throughput. Therefore the overall interference to the environment is reduced. By looking at the power level curve between the 150th and 170th second in figure 2.8 we see a stepwise reduction of the power level, showing the adaptation phase of Minstrel-Blues.

2-link Minstrel-Blues Experiments With our 2-link experiment we can analyze the impact of Minstrel-Blues on throughput while two interfering links are transmitting in parallel. The general setup is shown in ﬁgure 2.9. There are four Nanostations in adhoc mode on channel 40. All of them are connected to a measurement server via Ethernet to collect the PCAP and RegMon trace ﬁles.

Deliverable D.4.7 28 2. MinstrelBlues 2.5. Main Achievements and Challenges

Figure 2.7: Throughput and mac80211 states over time with TPC enabled/disabled.

In order the analyse Minstrel-Blues performance on this 2-link setup, we have sent four Nanostations to our CONFINE partner UPC in Barcelona. By having those nodes deployed in the CONFINE testbed we have the advantage to run Minstrel-Blues in a real WiFi network, where usual interference patterns arise from neighbouring 5GHz links. The setup of those four nodes, shown in ﬁgure 2.9, with a stable remote access is quite challenging and not working yet. Beside the usual CONFINE remote tests where researchers use the virtualized CONFINE infrastructure, our setup is a special case. In order to adjust and analyze Minstrel-Blues at the MAC-layer we need root access to our nodes. We are still working to get a proper setup deployed, so that we can run the last 2-link experiments.

2.5. Main Achievements and Challenges

Our main contributions are threefold: Fisrt we have developed a new open-source MAC-layer monitoring tool RegMon for Atheros ath9k Linux drivers. This enables the community to analyze the wireless medium at layer 2 and hence enables us to investigate in spacial reuse potentials by our TPC algorithm as we use RegMon to measure interference between links, second we have developed and releases a Linux based application to control a digital attenuator for proper wireless validation experiments and third we have implemented our joint rate and power controller Minstrel-Blues to current IEEE 802.11n devices driven by Linux OpenWrt and shown its potential to reduce transmit power while providing the same link throughput.

Deliverable D.4.7 29 2.5. Main Achievements and Challenges 2. MinstrelBlues

Figure 2.8: Transmitpower and SNR over time with TPC enabled/disabled

Our contribution is the new ath9k MAC-layer monitoring tool RegMon released as open-source under Github (https://github.com/thuehn/RegMon). All instructions to build, compile, run and use RegMon for own research purposes are provided. This enables everyone to analyze current Atheros WiFi hardware states at the PHY and MAC layer. As there was no Linux application for the attenuator we developed a new control tool. This command line application is released as open-source on Github (https://github.com/thuehn/ Labbrick_Digital_Attenuator) with instructions how to build and use it. This enables everyone with a Vaunix attenuator to run WiFi experiments by using a Linux environment. To work with multi-antenna devices multiple attenuators will be needed. By now our application only supports one attenuator at a time and needs to be extended for multi device support either by running it multiple time with the attenuators id, or by multi-threading the application. With the access to current IEEE 802.11n hardware from CONFINE, we were able to implement and test our joint rate and power controller Minstrel-Blues on those devices. Our experiments show that a dynamic adaptation of transmission power per packet is able to reduce the power level while providing the same throughput, hence proofs to be an essential contribution to reduce overall interference within today WiFi networks, which still use a static and rather high power level for all packet transmission. An additional experiment in CONFINEs real world environment is pending right now. For this experiment we will use four Ubiquiti Nanostation M5 which are identical equipped as our desktop experiment versions in the CONFINE testbed at UPC Barcelona remotely. We expect to show the

Deliverable D.4.7 30 2. MinstrelBlues 2.6. Conclusions

Figure 2.9: Minstrel-Blues Setup at UPC spacial reuse potential by using Minstrel-Blues. In addition to our current research code base of Minstrel-Blue we got to get several patches accepted and integrated into the mainline Linux kernel (https://git.kernel.org/cgit/linux/ kernel/git/torvalds/linux.git/log/?qt=grep&q=Thomas+Huehn). This includes a parsable version of the rate statistics for better evaluation of our experiments and several MAC layer restructure work to add TPC to the Linux kernel and hence get Minstrel-Blues upstream. We made our current Minstrel-Blues patch series also available to the public as open-source repository at Github (https://github.com/thuehn/Minstrel-Blues).

2.6. Conclusions

In Minstrel-Blues we developed and implemented an joint rate and power controller on current CON- FINE IEEE 802.11n routers. We performed several experiments that proof the ability to adapt transmit power level per packet. This decreases the needed power level while the link throughput stays constant. Therefore dynamic transmit power control with off-the-shelf WiFi hardware is a feasible and important part to reduce the growing interference level by efficient resource usage. As soon as we got all patches accepted upstream into the Linux kernel the overall network sum throughput will increase in today’s WiFi network deployments. In addition to our dynamic power controller we have implemented a new PHY and MAC layer monitoring tool RegMon and released it to the community. This new monitoring tool enables researchers to analyze those lower network layers in a fine-grained manner, especially the impact of interferences. With the support from the CONFINE project and community, we made a significant contribution to increase the efficiency of WiFi resource usage. One of the most promising future steps is the integration of Minstrel-Blues with the routing layer in mesh networks via cross-layer communication. The CONFINE research project OLSRv2 seems to fit very well to investigate the potentials of such an approach.

Deliverable D.4.7 31 3. Reﬂection - Enhancing Reﬂection and Self-Determination in Community Networking

This chapter discusses the project Enhancing Reflection and Self-Determination in Community Net- working (Reflection). One of the main outcomes of the project is the Network Characterisation Dae- mon (NCD), a piece of software that provides users of CMNs with an interactive tool to monitor, evaluate and fine-tune their network nodes.

3.1. Introduction

Community Networks (CNs) are IP-based networks designed, built, operated and maintained by communities of individuals that join together and cooperate to satisfy their telecommunication needs. They consist of distributed and decentralised network devices -hooked up via wired and wireless links- that interconnect computing systems, service providers, content repositories, end user devices, ..., handling the traffic between them. The sizes of CNs range from a few nodes to the tens of thou- sands. The main difference between the CNs paradigm and the traditional commercial Internet Services Providers (ISPs) is that end users are not mere consumers, but active contributors and stakeholders of the infrastructure. This empowerment comes along with rights and duties: having a voice for decision-making, the obligation by some sort of network license or agreement and, to a limited extent, the freedom to audit and participate in the control and management of the network resources and infrastructure. There are, however, several obstacles that significantly limit this freedom. Leaving aside CN issues like e.g. participants’ organisation, technical and legal aspects of the network deployment, this document focuses particularly on the day-by-day usage, monitoring and maintenance of the CN. For such an important task, there is a lack for convenient tools to help end users evaluating and understanding the state of the CN as a whole (or, at least, the part of the network around them, which plays the most important role in the perceived performance and quality of experience). Despite the existence of many networking tools to inspect particular characteristics of Metropolitan Area Networks (MANs), they are tailored for use cases that significantly differ from those of CNs. Typically, MANs are owned or controlled by a single or few entities. There, persons in charge of the network administration are well-skilled, able to perform complex evaluation tasks and have comprehensive control over all core components. This is not necessarily the case for an average CN user, whose boundaries for a comprehensive control are given by the nodes he owns. Furthermore, his/her skills can range between those of customers of a traditional ISP (i.e. being able to deal with an intuitive web portal on the home router) and those that have a solid technical background. To address the shortcomings of convenient tools for CNs, their members have created pieces of software to e.g. visualise the network topology and evaluate the links’ quality and bandwidth capacities. However, these tools do not provide sufficient feedback about the performance of individual links and end-to-end (e2e) routes, nor illustrate the chosen path and the links involved when traffic is sent towards a given destination. This lack of information and control significantly hinders the potential of

32 3. Reﬂection 3.2. Related work these networks and their routing protocols (RPs) to simultaneously adapt the route selection to both (i) temporary or long-term topology characteristics and (ii) user- or application-speciﬁc priorities.

3.1.1. Objectives and Vision

The envisioned benefits of the NCD for CN users shall be illustrated by an example: a non-expert CN user is trying to obtain the best possible downlink connection to a server outside of the CN, reachable only via Gateways (GWs) or proxies located several hops away. By looking at the current network topology, shown on the web interface of the home router (the CN node), the user can learn about the topology of the CN, the capacity of the links, etc. and actively measure network performance in real time. As a result, the user can take actions to improve his/her network usage experience, by configuring the home router to apply different routing policies that result in better performance. This way, optimal trade-offs between path delay, bandwidth and packet loss can be selected for different applications (voice over IP, large file downloads, etc.). The NCD contributes to fill the gap between the CMN (consisting of devices running the operating system (OS) and the RP, which provide most of their information via the command line, and the average CMN user, who is eager to interact with the network nodes using the mouse on a graphical environment. This is achieved by implementing a web-based graphical interface that allows the user to learn about the current network status, actively run performance tests and modify the routing policies of his/her network devices. The interface is built around a graph where the network topology, links’ quality, etc. are displayed from a user-centric point of view. Additionally, the NCD provides a framework for experimentation that allows to systematically run sets of network tests, displaying the results in the graphical web interface itself. This capability is exploited and discussed later.

3.2. Related work

Various tools exist for CN providing a global overview of the network topology as well as detailed information about existing links and nodes. Some of them take the information from a rather static database (Guifi Network Map, WiND, Nodeshot); some retrieve data from the nodes dynamically via Simple Network Management Protocol (SNMP) (Freimap) and others use a combination of both (Nodewatcher). A downside of these tools is that they require some sort of centralised infrastructure, introducing single points of failure. Moreover, advanced knowledge on systems administration and networking is required to deploy them. Another type of mapping tools are decentralised monitoring services, which are often integrated in Mesh Node System (MNS) firmwares and are accessed via the nodes’ web interface (FreiFunk, Gracia` Sense Fils (GSF), qMp, Lugro-mesh, Commotion, etc.). These tools mostly rely on the topology information locally available at the RP (optionally, they enhance the representation with street map data from external services like OpenStreetMap). These tools can show nodes and links that are known to the running RP but cannot assist in any centralised management task like node registration, address assignment or representation of planned links. However, they are instantly available to the users and provide a live snapshot of the topology. The existing tools lack means to link a given CN topology with the experienced performance at a particular location in the network. Besides, they do not assist at all in fine-tuning the RP to improve the network experience. There, tools like traceroute, tracepath, My TraceRoute (MTR), etc. are only of partial help because they only provide unidirectional e2e routes, leaving the impression to the user

Deliverable D.4.7 33 3.3. Social and Technological context 3. Reﬂection that the return path is symmetric. This is often not the case for routes in CMNs where forward and return paths are chosen independently.

3.3. Social and Technological context

In this section we introduce CMNs as a specific case inside the Guifi.net1 network ecosystem. Later, the qMp Mesh Node System and essential concepts of the BMX6 RP are discussed, as both have been leveraged in our work to integrate user-specific routing and network interaction into the NCD.

3.3.1. Community Mesh Networks in the Guiﬁ.net environment

The NCD tool is based on Routek’s field experience in deploying CMNs that are integrated in Guifi.net. This very heterogeneous CN is mostly built up with nodes that use the Border Gateway Protocol (BGP) and Open Shortest Path First (OSPF) RPs. Most of these nodes are connected by wireless links operating in infrastructure mode (Access Point (AP)/Client). However, in the last few years, some parts of Guifi.net operate as a qMp CMN in ad-hoc mode and using BMX6 as the main RP. The majority of these CMNs are geographically located in or close to Barcelona (Catalunya), but operate independently. Interconnection between these mesh islands exists via the infrastructure links of the Guifi.net network. CMNs in Guifi.net are either managed by the users, who own the nodes and participate in their maintenance, or by a few volunteer network administrators on which users trust. This means that physical or remote administrative access may not be immediately obtained (if at all) on certain nodes by the rest of the users. Furthermore, very different versions of the qMp MNS firmware are in use in these CMNs, as some nodes are not upgraded on a regular basis after their deployment. One of the goals related to the development of the NCD software is to install it on all the nodes of a real-life CN to test it and perform experiments with nodes in a production environment. To do so, administrative access to the nodes is required to update the MNS firmwares and install the required packages. This limits the number of candidate CMNs where to deploy the NCD in a controlled way.

3.3.2. Quick Mesh Project (qMp) qMp is an operating system for wireless routers aiming to facilitate the quick and easy deployment of CMNs. It is based on OpenWrt2, the de facto Linux derivative used for building free and open-source software (FOSS) firmwares for wireless network devices. The qMp firmware provides a web interface and a series of tools and scripts that act as an abstraction layer between the user and the underlying operating system and the RP, making configuration and management of CMN nodes easy. In CMNs based on qMp, all the nodes run BMX6 as their main routing protocol (and, optionally, some nodes may run another protocol to allow interconnection with other CN segments). The qMp MNS has been selected for integrating the interaction tools developed in this work due to its active usage and development by local network communities and its already existing native support for the BMX6 routing protocol.

1Guiﬁ.net: http://www.guifi.net 2OpenWrt: http://www.openwrt.org

Deliverable D.4.7 34 3. Reﬂection 3.4. The NCD: a characterisation and interaction tool for CMNs

3.3.3. BMX6 routing protocol

BMX6 is a distance-vector routing protocol for Linux-based operating systems. It has been designed and discussed in the context of the very dynamic wireless Community Networks, where the quality of the links established between nodes can greatly vary within a short period of time and multiple paths can be found at any time to transmit packets from one node to another. BMX6 is a lightweight and efﬁcient application, so it can be run on many of the routers typically used in Community Networks. These embedded devices are very heterogeneous but, in general terms, they have limited resources (a 400 MHz SoC CPU and 32 MB of RAM are typical hardware speciﬁcations, although faster devices with 64 MB of RAM or more are recently becoming the standard).

3.3.3.1. BMX6 general operation concepts

BMX6 uses User Datagram Protocol (UDP) broadcast messages to exchange link, node, and path- discovery messages between neighbouring nodes and, by re-broadcasting requested messages on demand, propagate globally-relevant information to all network nodes. Unlike traditional routing protocols, a BMX6 node (e.g. a wireless router) is not identified by its primary IP address, but by two ID values of global scope, being (i) a permanent ID that identifies a particular router at any time and (ii) a description ID that is generated by the current set of configuration parameters of a router at a given moment in time. This set of configuration parameters contains, among others, its global permanent ID, a sequence number, the announced address ranges reachable via this router, and the specification and parametrisation of a metric function that defines how forwarding routes (next hops) towards the announced address ranges should be selected and how the path metric value propagated via routing updates should be calculated by other nodes of the network. The description ID is simply given by the SHA1 hash of each router’s configuration. Routing updates in BMX6 contain 3 fields. A path-metric value, a sequence number, and, instead of a destination address or network, a node-configuration reference such as the description ID. Based on the routing update messages originated from a given node and received via a given link, the path- metric value of that update and the metric function defined via the referenced description of the originating node, an updated end-to-end path metric is calculated. This way, once the description - referenced by a routing update and received via one or several links (i.e. neighbours)- is resolved, the best next hop (i.e. the route) for forwarding data packets to the originating node is given by the link with the best path metric value. This value is also used as metric when re-broadcasting the received routing update for further e2e path propagation.

3.4. The NCD: a characterisation and interaction tool for CMNs

3.4.1. General concepts

The NCD is conceived as a decentralised software running independently on each node in the CMN. It consists of a web-based graphical interface (the NCui, Sec. 3.4.2.2), which integrates with the MNS’s web interface, to handle the interaction with the end user. This interface retrieves and modiﬁes status data and conﬁguration from both the local and the other nodes via the NCD daemon (called lunced). The lunced daemon runs in the background and is able to ask and send information and commands to the local OS and RP (Section 3.4.2.1). Additionally, it implements mechanisms to exchange information with other NCD daemons running on the other nodes of the CMN (Section 3.4.2.3).

Deliverable D.4.7 35 3.4. The NCD: a characterisation and interaction tool for CMNs 3. Reﬂection

3.4.1.1. Requirements

The NCD software is designed to be installed and run on CMN routing devices using a MNS based on OpenWrt 14.07 or newer (like the stable version of qMp). The lunced daemon is very lightweight; its hardware requirements are not higher than those of the MNS itself and only 20 kB of free disk space and around 100 kB of free RAM are needed to install and run it. The daemon is a stand-alone program, but depends on the following OpenWrt packages: libc, lua, liblua, libubox-lua, libubus-lua, ubus and iperf3 (most of them are already included in the MNS ﬁrmware by default). The NCui takes approximately 310 kB of free disk space for the installation, but since all the data manipulation occurs on the client’s web browser side, no extra resources are needed on the router. The web browser must have JavaScript capabilities enabled to properly display the graphical web interface.

3.4.2. The NCD: architecture, components and considerations

The NCD consists of two software components. On the highest abstraction layer, the NCui web-based tool shows information about the network, evaluates performance parameters and provides interaction mechanisms to change routing metrics, etc. The interface integrates with the daemon on the lower abstraction layer called lunced, the NCD daemon (NCd) itself. lunced manages the interaction with the local device components and the remote network nodes. The architecture of the NCD is depicted in Figure 3.1, showing the external components lunced is surrounded by.

Figure 3.1: Architecture of the network characterisation daeomon

Deliverable D.4.7 36 3. Reﬂection 3.4. The NCD: a characterisation and interaction tool for CMNs

3.4.2.1. The NCD daemon (lunced)

The core of the NCD is the lunced daemon (which stands for Lua Network Characterisation Daemon. It interacts with the other system components and enables the communication between them. It has a modular design, where plug-ins provide specific interaction functionalities (e.g. with the routing algorithm). This interaction is divided in three blocks: user I/O, [mesh] network I/O and system I/O. User I/O in lunced comprehends a set of tools to allow the end user to interact with the NCD in three different ways so far: web interface (see Sec. 3.4.2.2), system log and command line. The system log interaction is possible via a specific plug-in that allows lunced to log certain system events (e.g. requests received from remote NCD-capable nodes). Command-line interaction is based on the interface provided by OpenWrt Ubus. The NCD is actually designed to perform the communication between the NCui and the NCd via Ubus. Backwards compatibility has been taken into account in the NCD since the beginning. Once a minimal set of functionalities have been defined and tested, mechanisms for interoperability with old revisions were added to the newer versions of the software. In fact, the NCd is able to interact with nodes of the CMN not even running the NCD software. This issue has been addressed with special interest because, from our experience, different versions of the same MNS tend to coexist in a CMN and nodes are not, in general, updated in a regular manner as new versions of the firmware are published. The modularity of the NCd allows room for all types of plug-ins and new developments. For instance, if a new RP is to be added, only a plug-in needs to be created.

3.4.2.2. The NCD user interface

The NCui provides CN end users with a visual tool to monitor the network and perform administration tasks accordingly in order to improve their usage experience. It features a graph showing the network nodes and the links connecting them. The information is displayed in a user-centric approach (which makes sense in the context of a mesh network with BMX6, as the RP is not aware of the complete network topology). Therefore, at the initial stage, only the information locally available is displayed on the interface (the local node and the links to the neighbours). As the user asks for more information, the NCD daemon asks the neighbouring nodes and nodes further hops away, about their details and status. This way, the user can gradually learn about a bigger part of the network until the whole CMN is shown. The process of discovering a whole CMN with the NCui can be seen in Figure 3.2. Fig. 3.2a shows the initial view, with the local node in the middle and its three neighbours around it. In Fig. 3.2b, the graph is centred on the node on the right after its list of neighbours has been requested. No new nodes are added, but links previously unknown have appeared. The discovery process is repeated until all the nodes appear on the graph (Fig. 3.2c depicts the whole CMN). Finally, Fig. 3.2d uses a colour palette (blue → green → yellow → orange → red) to paint the links in function of their measured quality. In addition to the visual information displayed by the graph, the NCui also provides text-based data about the nodes, the routing algorithm, paths, etc. by means of a sidebar and ﬂoating HTML divs when needed. Figure 3.3 shows the BMX6 path discovery between two nodes in the mesh network. The user can use the NCui to perform network performance tests between their node and another node in the CMN. For instance, latency and bandwidth can be evaluated between the user’s router and different gateway nodes that offer an Internet up-link. In this case, the results are displayed in tables and graphs on the web interface itself, and can offer the user valuable information for selecting

Deliverable D.4.7 37 3.4. The NCD: a characterisation and interaction tool for CMNs 3. Reﬂection

(a) Local node in the center, with all its neighbours (b) The second node’s neighbours and new links and links. are shown.

(c) All the mesh nodes are shown, and also the links (d) The network links are coloured as a function of connecting them. their quality.

Figure 3.2: Discovery of the nodes in a mesh network and the links’ properties in four steps using the NCui. a more appropriate gateway, or for adjusting certain parameters of the RP to improve the network quality of experience.

3.4.2.3. Inter-NCD communication

One of the key features of the NCD is that the lunced daemon can exchange information by sending and receiving JSON queries over HTTP. To do so, a special function is used to forward local queries to other nodes running the NCd. This occurs when the user manually requests information about a remote node or at timed intervals, but only when the NCui is in use. One of the current shortfalls of the NCd implementation is that all the requests a node receives from other nodes in the mesh are executed, no matter where they come from. The next development effort will take this into account to at least provide a conﬁguration option to instruct lunced to only attend read-only requests and discard the ones intended at changing the node’s settings. Ideally, an authentication mechanism should be implemented (for example, by exchanging RSA keys and sending write commands securely via SSH).

Deliverable D.4.7 38 3. Reﬂection 3.5. Achievements and Challenges

Figure 3.3: Screenshot of the NCui showing the mesh graph and highlighting the path from the far right node to the one on the far left.

3.4.3. NCD source code and OpenWrt package for dissemination

The source code of the NCD is freely available under a GPL licence and can be run on the current (as of February 2015) OpenWrt stable version. Packages for OpenWrt are also provided with the development releases of qMp and it is planned to be included in this ﬁrmware’s next stable release.

3.5. Achievements and Challenges

Routek’s interest in participating in CONFINE’s 2nd Open Call (OC2) consists of two elements. On one hand there is the company’s engagement with CMNs and in particular with the Guiﬁ.net CN. On the other hand is the interest in developing a piece of software to improve the company’s current applications and offer the newly created application to current and new customers, in order to be able to offer better products and services.

3.6. Achievements and challenges for Community Mesh Networks

The main achievement after Routek’s participation in CONFINE’s OC2 is the integration of the NCD in the qMp MNS ﬁrmware. Once this tool is published with the next stable version of qMp, end users will be able to learn about their CMN and better understand how it performs. For instance, information about which are the main nodes and links of the network, etc. will be accessible for them in a way that had not been available before. This will facilitate the day-by-day tasks of monitoring and maintaining the network and, in turn, lower the entry barrier for new users, as the process to use and manage the network is easier and more user friendly. One of the biggest challenges regarding the integration of the NCD with the qMp MNS ﬁrmware is managing the feedback provided by the users. The active participants of CNs usually enjoy testing new pieces of software and are eager to send comments, suggestions and also complaints to the

Deliverable D.4.7 39 3.6. Achievements and challenges for Community Mesh Networks 3. Reﬂection developers via the available contact mechanisms (mainly the mailing lists). It is important to capitalise on all this feedback. This is done by consider the feedback to improve the tool, by ﬁxing the bugs that may eventually appear, satisfying the requests for new features, etc. and, as much as possible, by helping users to contribute to the code by incorporating submitted patches.

3.6.1. Achievements and challenges for commercial exploitation

The NCD is based on the previously existing tools and pieces of software already in use by the company. Thanks to the resources received, the company has been able to greatly improve them and test them in both the WiBed testbed environment and a production CMN. Having the NCD in a production-ready status is an important milestone in Routek’s development plans. In network deployments where the mesh network technology is used but, unlike in CNs, centralised control and network administration is required, this tool comes in hand to get both an overall live view of the network status at a glimpse and to obtain the details of particular nodes or links. Therefore, the combination of ﬂexibility, failure tolerance and resiliency of wireless mesh networks and the features provided by the NCD help the company offering competitive network solutions for our customers.

3.6.2. Conclusions

The participation in CONFINE’s OC2 has been very positive for Routek, as it has provided the company with resources for researching and development on topics related to our sphere of activity (mesh networks, dynamic routing protocols, network management tools, distributed systems, etc.). Addi- tionally, the opportunity to use the WiBed testbed has been of great help to test our software in a controlled environment very representative of the target site of typical network deployments developed by the company. First, the development of the NCD has provided both the company and the end users of CMNs with a tool to better understand, learn about and administer wireless mesh networks in a novel way. Second, the resources for research have given the opportunity to investigate more in the topic of dynamic routing protocols (especially BMX6), improving this open source software that is a key piece in Routek’s activity. Third, and ﬁnal, the experimentation performed during the project has lead to a better understanding of the behaviour of production wireless mesh networks. This information is essential in order to improve current and future network deployments, either in the context of CNs or private/commercial networks.

Deliverable D.4.7 40 4. Blessing of The Commons: Improving Radio Resource Utilization Efﬁciency in Community Wireless Networks

4.1. Motivation

We start by making the argument that wireless community network growth1 is highly impacted by the effort necessary for individuals to support the network by adding and maintaining an additional node. WCNs can only grow if the locally available cognitive surplus (time and skills) is high enough to sustain those required efforts. Thus, in such absence of capital investments, the metric that we need to maximize is performance vs effort. Currently, the effort mentioned above seems lower bounded by installation and maintenance cost of a rooftop node. In most relevant contexts (urban as well as rural) a manual highly accurate configuration of directional-antenna links and constant radio network planning and optimization by experts is necessary. Unfortunately, that is something that upstarting wireless community initiatives rarely manage to achieve on their own, as of today. The key question is if innovations at the technological level can significantly contribute to an improvement of this situation. In the second half of this report we propose, describe and demonstrate small modifications to the most popular (IEEE802.11n-AtherosChipset-GNULinux) software stack used in current WCN contexts. The rationale behind all of these modifications is: introducing common-radio- resource-cost-transparency within the network to increase cooperation gain and improve the networks capability to self-regulate and self-optimize. But before we go into the details of how to achieve utilization transparency, we must fully understand what radio resource cost really is in wireless mesh networks. Therefore, the first half of this report is dedicated to understanding the detrimental effects of interference in IEEE802.11n-ath9k based networks that are using an OFDM physical layer and listen before talk (CSMA/CA) medium sharing mechanisms. In the context of such a PHY/MAC stack, one can define interference to be the occurrence of:

Signal power being received at antenna input of a single node is comprised of signiﬁcant energy components from MULTIPLE sources which are not fully orthogonalized in signal space by demodulation.

According to theory that is equivalent to any overlap in time-frequency plane before OFDM demodulation. Surprisingly, very little empirical measurement data has been published that allows to quantify the cost that such overlap has in current IEEE802.11 OFDM chipset implementations in terms of reduced frame error ratios (FER). We want to better understand how to build high performance networks without using spatial link isolation techniques such as rooftop installation to minimize that overlap2 - so we ﬁrst need to quantify

1This growth, for example measured in total number of nodes divided by total area covered has not increased by the same rate that the HW prices have fallen during the last 10 years. 2Otherwise the wireless network becomes interference limited very fast with growing spatial density as the dominant cause of error is not thermal noise in the receive chain, but power components from other sources than the intended

41 4.2. Part I - Empirical Model of Interference in IEEE802.11 4. BTC the cost of interference on FER in such non-spatially isolated links in order to be later able to assign it the correct cost. Note, that we are not arguing that capacity scaling is achieved by utilization transparency. But we think that anything that helps to understand and minimize the detrimental effects of interference certainly helps the general cause of wireless community networks.

4.2. Part I - Empirical Model of Interference in IEEE802.11

We present methodology, experiment configuration and final results from a set of measurements which provide detailed insight on the characteristics of current receiver implementations in terms of their capability to successfully decode OFDM-frames which are overlapped in time by one or more lower-power frames received on the same carrier frequency (commonly referred to as physical layer capture). In order to do that we devised a measurement method that allows to reliably reproduce interfering signals with very high accuracy timing-relations between interfering OFDM frames. Subsequently, we also adapt the simulation model of NS-3 in order to reflect the observed capture capabilities when simulating IEEE802.11 OFDM based wireless networks.

4.2.1. Introduction

Event based network simulation tools are an integral part of wireless ad-hoc networking research. They allow us to predict network performance very cost-efficiently and offer full repeatability which is hard to achieve via large-scale measurement campaigns. Nevertheless, the credibility of network simulation results remains low as long as important model assumptions are not validated by empirical measurements. This seems to be the case for receiver error models in IEEE802.11-type network simulation – and the respective interference model assumptions in particular. Possible explanations for this gap are: first, the involved effects that cause reception errors under interference (i.e. when signals at the receiver input are not perfectly orthogonal in time, frequency or code) are more complex in nature than those effects that determine reception performance in the presence of thermal noise only. Second, characterizing and modeling an existing receiver implementation in this regard requires to reliably generate test signals that reflect a well defined power and timing relation between interfering signals at the antenna port. However, during our previous work in the field of VANETs (based on a variant of IEEE802.11a) we repeatedly made the observation that system performance is highly impacted by the way transceivers can cope with interference (multiple frames carrying individual cooperative awareness messages from neighboring vehicles overlapping in time at a single receiver input). That is mainly caused by the lack of any hierarchical protocol structure to keep the effects of interference under control (like it is done with the help frequency division duplex and/or central radio resource control in cellular networks). In many relevant scenarios, expected node densities may cause the system performance to be dominated by interference-limitations and therefore, the question arose repeatedly if mesh protocols that have been successfully simulated also work as expected under real life conditions. Large-scale field test would be necessary to answer this question exhaustively, but are very expensive and time consuming

transmitter. In such contexts the total network performance cannot be improved by increasing individual transmission power).

Deliverable D.4.7 42 4. BTC 4.2. Part I - Empirical Model of Interference in IEEE802.11 to maintain. If done at all, field tests often consist only of a few nodes, and they cannot be replicated easily. In order to remove uncertainty in that respect (and build an interference mode that is well supported by measurements) we have created a system that allows to synthesize a radio frequency signal that contains fully standard compliant 802.11 OFDM frames overlapping in time which is then fed to the device-under-test (DUT) receiver. The key element of this measurement approach is using software- defined radio (SDR) elements in the signal generation stage in order to being able to replicate timing relations with sub-microsecond precision. This principle allows to do things that would be infeasible using standard off-the-shelf transceivers as transmitters: we are able to accurately reproduce signals that reflect interference events known to occur regularly in CSMA/CA-based medium access networks (e.g. hidden terminal situations or MAC layer collisions). The contribution of this paper is as follows: • We provide a short review of the interference model that is currently used in the network simulator NS-33. • We present a method to synthetically generate RF signals that reproduce interference events in IEEE802.11-based OFDM networks. • We show how this approach can be used to verify model assumptions, such as the presence and characteristics of the physical layer capture effect. • We adapt the current interference model of NS-3 to match the results for one particular receiver implementation.

4.2.2. Related Work

By default, NS-3 assumes a receiver error model which describes the reception performance in the presence of Additive White Gaussian Noise (AWGN) and this model has been empirically validated in [20,21]. But only very few measurement-based papers on the subject of reception errors caused by same-channel frame overlaps in OFDM-based IEEE802.11 networks have been published yet. The single most relevant work seems to be [22]. Also, the interference model in IEEE802.11b-based (direct sequence spread spectrum) networks has been analyzed in [23]. However, the measurement results presented in those related works have been gathered using standard IEEE802.11 transceivers to generate the test signal and therefore do not provide the timing accuracy for analyzing the phenomena that are the main focus in our work. So, since no measurement data-set with a higher level of detail seems to be available publicly up to now – all simulation-based contributions in the context of detailed interference modeling and capture effect analysis build their assumptions on the observations presented in [22] and [23]. Such is the case for [24], [25] and [26] where the authors are able to pinpoint the relevance of the capture effect and show that subtle changes to the interference model can cause significant deviations in terms of system level performance. Also, a number of theoretical-only contributions to the field of interference modeling in CSMA/CA networks are found in [27] and [28]. Interference constraints are expected to determine system performance specifically in high-density scenarios and that motivated [29] as well as [30] to focus on that subject. Our motivation is to shed more light onto the true cost of interference for receive frame error ratios (FER) and increasing the accuracy of NS-3 simulation results in interference-limited scenarios. To the best of our knowledge, no investigations at this level of detail have been published yet.

3http://www.nsnam.org

Deliverable D.4.7 43 4.2. Part I - Empirical Model of Interference in IEEE802.11 4. BTC

4.2.3. The Current NS-3 Interference Model

The NS-3 interference model currently implements a straightforward concept of additive interference power accumulation: If a frame is overlapped in time by other frames at a common receiver antenna input, then the reception powers of interfering frames are treated as additional thermal noise contributions during the overlapping segment of the current frame. This model was introduced in the simulator YANS [31] and later adopted in NS-3.

4.2.3.1. The NS-3 interference helper

(a) Frame B PLCP Header PLCP Payload

Frame A PLCP Header PLCP Payload

(b)

Frame B Power

Frame A

time Noise floor NI changes

Signal Frame A

Chunks: c1 c2 c3

relevant noise and Frame B interference level Noise+ Interf.

relevant NI change time

Figure 4.1: The INTERFERENCEHELPER splits frames into constant SNIR chunks and the current NS-3 interference model treats overlapping frames simply as if they were temporary noise contributions.

Assume Frame A is overlapped by Frame B as depicted in Figure 4.1. Internally, NS-3 employs a so-called INTERFERENCEHELPER function to loop over all relevant network interface changes (NI changes) during the duration of Frame A, which are the start and end events of Frame B in this case. At each time of such an NI change event, a new chunk of Frame A is deﬁned to start. An additional split is introduced between preamble and payload parts of the frame in consideration, since the payload may be encoded with a higher modulation and coding rate than the preamble (which is always encoded with BPSK and a half-rate convolutional code).

For each of the resulting chunks (c1, c2, c3), the corresponding signal-to-the noise-and-interference ratio (SNIR) is calculated by adding the received power from interfering frames to the denominator (i.e. treating them as additional AWGN contributions). In the following step, an empirically validated ERRORRATEMODEL is used to map these SNIR values per chunk to the corresponding coded bit- error probabilities pb,i for each chunk i. The chunk success rate pcsr,i for a chunk of length n bits is deﬁned as: n pcsr,i = (1 − pb,i)

Deliverable D.4.7 44 4. BTC 4.2. Part I - Empirical Model of Interference in IEEE802.11

and the overall packet success rate Psucc is then obtained by multiplication of success rates of all relevant chunks. Finally the inverse gives the packet error rate Perr:

Psucc = pcsr,1 · pcsr,2 · ... · pcsr,n

Perr = 1 − Psucc

The following key assumptions are adopted when modeling interference in this way: • Interference is treated as a temporary increase of the thermal noise floor at the receiver input. Thus, the assumption is that interfering OFDM signals have exactly the same detrimental effect on the reception process as Additive White Gaussian Noise (AWGN). This is a valid starting point but has not yet been verified rigorously in the case of interfering IEEE802.11 frames. • Individual chunks success ratios are treated as independent random experiments. That corresponds to the assumption that the forward error correction (FEC) – a standard convolutional block code – is unable able to mask error bursts. • Frames received with an absolute power below the so called ENERGYDETECTIONTHRESHOLD at the antenna input are ignored. This is done in order to reduce computational complexity since interference terms from very distant sources can be disregarded right away. By default, this threshold is set to -96dBm – which is slightly below the noise floor of -97dBm (using the default receiver noise figure of 7dB).

High-power frame first High-power frame

Low-power frame

High-power frame last High-power frame

Low-power frame

- frame length 0 + frame length t (adjacent) (exact overlap) (adjacent)

Arrival delay of the high-power frame

Figure 4.2: Message-in-message constellations: We consider two equally sized frames with different power levels, separated by a deﬁned signal-to-interference ratio (SIR). The difference of arrival times is speciﬁed by the arrival delay of the high-power frame.

One way to characterize the capture effect (or lack thereof in the current model) is to run two-frame- overlap simulation experiments as depicted in Figure 4.2. This replicates a canonical hidden node scenario, two transmitters are unable to sense each other to coordinate medium access and their signals overlap at a common receiver node. By varying the start time of both frames and keeping track of the frame error ratio of the high-power frame we can generate Figure 4.3. It reﬂects the standard NS-3 model behavior for several SIR ratios (the relation between high-power and low-power frame). Clearly, this Figure demonstrates the effect of interference power accumulation, as the error ratio for decoding the high-power frame starts to

Deliverable D.4.7 45 4.2. Part I - Empirical Model of Interference in IEEE802.11 4. BTC

0.8 SIR = 5dB SIR = 6dB 0.6 SIR = 7dB SIR = 8dB 0.4 SIR = 12dB

0.2 FER of High−Power Frame

−1000 −800 −600 −400 −200 0 200 400 600 800 1000 Arrival−Delay of High−Power Frame [µs] Figure 4.3: Frame Error Ratio (FER) of the high-power frame for several SIR points observed in a simple two-frame-overlap simulation experiment (cf. Fig. 4.2). This ﬁgure also shows that the default NS-3 model does not contain any physical-layer capture model yet. gradually increase until complete overlap. It stays at FER=1 until the high-power frame is delayed by the length of the low-power frame (800µs), so both are completely non-overlapping in time again, thereby acknowledging that the physical layer capture effect is currently not part of the model. Even at very large SIR values the capability to resynchronize (in order to decode the frame which has a higher chance of successful decoding) is not part of the default interference model in NS-3.

4.2.4. Measurement Methodology

Our first approach to assess how closely the current NS-3 interference model assumptions match with observations from real world measurements was to precisely replicate the aforementioned two-frame- overlap experiment in our labs, using an implementation manufactured by Atheros as our device- under-test receiver unit. Our aim was to reproduce the frame-error-ratio curves depicted in Figure 4.3 – but this time based on measurement data. The main difficulty in doing this direct side-by-side comparison is to replicate the exact timing relations between interfering frames and reliably reproducing these for a large number of repetitions (10.000 in our case) to gather sufficient samples in order to come up with a statistically sound characterization. Note that the length of a single OFDM symbol within the frame is a few microseconds – and we want to be able to control during which OFDM symbol the interfering frames start to overlap because this allows us to confirm/refute preamble-dependent effects that have been reported in other measurement campaigns cited in the related work section. Using standard IEEE802.11 transceivers for transmit signal generation turned out to be an unsuitable approach in this case – simply because there is no way to reliably trigger the transmission time on such devices externally with an accuracy of under a few microseconds. In order to solve this problem we set out to build a testbed that instead uses two synchronized software- radio front-ends for up-converting two baseband signals that contain standard compliant IEEE802.11 OFDM frames slightly delayed between each other and superimpose them prior to feeding the signal to the DUT receiver.

Deliverable D.4.7 46 4. BTC 4.2. Part I - Empirical Model of Interference in IEEE802.11

4.2.4.1. Measurement System Description

Baseband Signal Device Monitoring USRP Encoder under test station Frontends

Figure 4.4: Conﬁguration used for measurement-based analysis of the reception performance under interference and the capture effect in particular. Therefore, two individually delayed IEEE802.11 OFDM frames are superimposed in radio frequency domain and fed to the receiver input of the DUT.

Figure 4.4 shows the basic layout of the resulting configuration which consists of the following elements: baseband signal encoder, radio frequency up-conversion units, additive signal superposition, the device-under-test and a monitoring station to record results. We make extensive use of the IEEE802.11 OFDM encoder that has been developed at Forschungszen- trum Telekommunikation Wien (FTW) and released under an open source license [32]. Sample-clocks at both SDR front ends are started with a common external timing strobe (PPS) and the individual baseband signals are then delayed by an appropriate number of samples at the signal encoder host. This allows us to set the delay between interfering frames with a resolution of 40 ns (at 25MS/s baseband rate). Note that local oscillators in the XCVR2450 direct up-conversion daughterboards are not correlated – thereby we mimic the real world situation of two independent transmitters as closely as possible. Both signals are then additively superimposed in radio frequency domain via a passive power combiner. This sum-signal is attenuated by fixed power attenuators to produce a given average target SNR and the final interference-signal is directly fed to a device-under-test – without being distorted by a wireless channel. Such an AWGN configuration is used since effects of time variability and frequency selectivity of the wireless channel shall be excluded in this measurement campaign. See Table 4.1 for more details. Figure 4.5 shows three exemplary baseband signal snapshots for different timing conditions after the additive signal combiner. These the two-frame-overlap experiments are repeated 10.000 times for each delay/SIR measurement point. Table 4.1 recapitulates the most relevant measurement parameters.

4.2.4.2. Transmit Signal Generation

Due to a limitation of the digital up-conversion device (Ettus USRPN210) we are not able to reproduce 20MHz wide transmission signals without loss of accuracy at the band edges. But characteristics observed in 10MHz measurements are expected to be fully representative also for 20MHz wide IEEE802.11a/g signals – since the only difference is that OFDM symbols are twice as long in IEEE802.11p (8µs instead of 4µs)- everything else in the encoding/decoding process is identical. Also, keep in mind that the we are only able to generate a single input stream - repeating the same procedure for a MIMO receiver requires considerably more effort and is planned as future work. Unfortunately, the XCVR2450 daughterboard from Ettus Research does not come with automatic LO- power suppression (DC compensation) capabilities, thus a signiﬁcant power component is emitted at 0Hz in the baseband signal as already documented in [32]. It was necessary to apply a shift in the

Deliverable D.4.7 47 4.2. Part I - Empirical Model of Interference in IEEE802.11 4. BTC

1 200 µs delayed

0.5 Sample Baseband Magnitude

0 0 200 400 600 800 1000 1200 1400 1600 Time [µs]

1 synchronized start

0.5 Sample Baseband Magnitude

0 0 200 400 600 800 1000 1200 1400 1600 Time [µs]

1 200 µs advanced

0.5 Sample Baseband Magnitude

0 0 200 400 600 800 1000 1200 1400 1600 Time [µs] Figure 4.5: The baseband signal after the passive power combiner for three exemplary delay settings (down-converted to digital baseband by a third USRPN210 used for monitoring purposes). Eventually, the measurement system allows to reproduce arrival time differences with an accuracy of less than 40 nanoseconds. digital domain in order to move the visible peak at the center frequency out of the desired signal bandwidth. Compare left and right parts of the power spectrum measurement plot in Figure 4.6.

0Hz intermediate frequency 10MHz intermediate frequency

Figure 4.6: Setting a digital intermediate frequency of 10MHz was necessary in order to shift detrimental artifacts (the local oscillator power at DC) out of the designated frequency band.

4.2.5. Measurement Results

Figure 4.7 presents the ﬁnal outcome of the ﬁrst batch of measurements. Same as in the simulation- based Figure 4.3, the y-axis shows the frame error ratio of the higher-power frame and the x-axis depicts the arrival delay of the high-power frame with respect to the lower-power frame. The signal- to-interference ratio is noted in dB and increases from 1 to 5dB in this particular plot.

Deliverable D.4.7 48 4. BTC 4.2. Part I - Empirical Model of Interference in IEEE802.11

Table 4.1: Measurement Parameters

SDR Frontend Model Ettus USRPN210 Daughterboard Model XCVR2450 Tuning Range 2.4-2.5GHz, 4.9-5.9GHz Reproducible Dynamic Range -100 dBm ...-20 dBm Local Oscillator Type internal TCXO (2.5ppm) DAC resolution 2x16 bits Baseband Data-Rate 800 Mbit/s (single stream) Baseband Bandwidth 25 MHz Interpolation Factor 2.5 Effective Signal-Bandwidth 10 MHz TX Carrier-frequency 5.88 GHz (Channel 176) MPDU Size 500 bytes Modulation Rate QPSK R=0.5 Framelength 800 µs Transmit Interval 10 ms Channel model AWGN (time invariant) High-Power-Frame RSSI -80 dBm Low-Power-Frame RSSI -80 dBm ... -95 dBm Delay Stepsize 25 µs Repetitions per Delay-Step 10000

1 SIR = 1dB 0.8 SIR = 2dB SIR = 3dB 0.6 SIR = 4dB SIR = 8dB 0.4

0.2 FER of Higher−Power Frame 0

−1000 −800 −600 −400 −200 0 200 400 600 800 1000 Arrival−Delay of High−Power Frame [µs] Figure 4.7: Measured frame error ratio of the high-power frame for several SIR points. Comparing this ﬁgure with Fig. 4.3 shows a systematic difference of 4dB in required SIR.

By comparing Figure 4.7 and Figure 4.3 one important difference becomes immediately apparent: signal-to-interference ratios that lead to identical FER characteristics are systematically lower in the measurement – by exactly 4dB across all SIR points. This observation indicates strongly that interfering OFDM signals have in fact less detrimental impact on the successful reception statistics. Treating them exactly like AWGN leads to pessimistic over- estimation of the detrimental effect of OFDM interference. One hypothesis is that this is caused

Deliverable D.4.7 49 4.2. Part I - Empirical Model of Interference in IEEE802.11 4. BTC by the energy within the OFDM signal being non-uniformly distributed over the channel bandwidth (c.f. Figure 4.6).

0.8 SNIR: 8dB 0.6 SNIR: 9dB SNIR: 10dB 0.4 SNIR: 11dB SNIR: 12dB

0.2 FER of Higher−Power Frame 0

−1000 −800 −600 −400 −200 0 200 400 600 800 1000 Arrival−Delay of High−Power Frame [µs] Figure 4.8: Measured frame error ratio of the high-power frame for higher SIR points. It shows that this particular receiver implementation is able to capture delayed high-power frames as soon as the SIR exceeds around 8dB and reaches capture probability close to one at SIR=12dB.

Another detail can be seen in Figure 4.7: this receiver implementation seems to be able to capture some high-power frames at SIR=8dB as the frame error ratio is starting to decrease slightly and is not strictly one anymore. In order to shed more light onto that phenomenon we repeat the measurements for higher SIR settings beyond 8dB, and plot the results in Figure 4.8. A number of interesting results can be drawn from Figure 4.8: • This particular receiver implementation is able to decode the high-power frames as soon as the SIR exceeds 8 dB. • No frame errors are observed across the whole overlap region as soon as the SIR exceeds 12 dB (dotted line). • Thus, the capture probability is not conﬁned to small arrival delay values (in contrast to what has been reported as preamble-capture effects in related work [22]). • Another subtle detail is the peak at 800 µs at SIR= 12 dB – it indicates that the frame-start detector is blind and misses the next frame if there is no inter-frame space of at least 25 µs after a previous frame has ended. • A non-intuitive observation is that there exists a capture probability plateau at high SIR values – the percentage of overlap does not seem to matter anymore. We expect this to be caused by timing estimation errors at the moment of resynchronization, and these estimation errors dominate over interference power contributions during the frame overlap phase.

4.2.6. Adaption of the NS-3 Interference Model

Our work was also motivated by the assumption that the interference model can have significant impact on final simulation results in the context of high-density networks. Therefore, we modified two components of the Wi-Fi model4 in order to reflect the capture effect as it has been observed in our measurements: 4The source code will be made available by inclusion in the next NS-3 release, which is pending.

Deliverable D.4.7 50 4. BTC 4.2. Part I - Empirical Model of Interference in IEEE802.11

StartRx

Current State? IDLE or Receiving (RX) CCA_BUSY Transmitting (TX) rxPower > SNIR > CaptureThreshold? EDThreshold? No No

Yes Yes

Drop current packet

SwitchToRx Cancel EndReceive event NotifyRxStart of current packet to InterferenceHelper Schedule EndReceive event Drop new packet

Figure 4.9: Flow chart for the processing of a start frame event in NS-3. The shaded part shows the additional processing for frame capture events. Without capture (default NS-3), a new frame that starts while in receive mode is always dropped.

First, the start frame processing in the physical layer class (YANSWIFIPHY) had to be extended in order to allow the switch to a new frame while the reception of another frame is ongoing. This switch is made under the condition that the receive power of the new frame exceeds the receive power of the old frame by a certain factor, which is specified by the capture threshold (see Figure 4.9). If the condition is met, the current frame is dropped and the reception of the new frame is started. The capture threshold was introduced as a new parameter of YANSWIFIPHY along with a flag to enable or disable packet capture. The second modification was necessary in the INTERFERENCEHELPER class, which tracks all frame start and end events over time, and is used at the end of the reception (end receive event) to calculate the SNIR for each chunk and derive the overall packet error. The existing INTERFERENCEHELPER can handle all frame overlap combinations, but it assumes that the first frame always starts in the absence of any interfering frame (i.e. channel is unoccupied). This was generalized in order to support the start of reception process during another frame being currently received, but without changing the power accumulation principle. The new parameters of YANSWIFIPHY do not enable capture by default, and the modifications of the INTERFERENCEHELPER do not change the non-capture case. Therefore our implementation is com- patible with the default NS-3 implementation. Changes become only visible, if capture is explicitly enabled.

4.2.7. Summary of Part I

Since interference experiments are difﬁcult to execute, very little validation data (supporting current interference model assumptions in NS-3) is publicly available. That motivated us to build a dedicated interference measurement testbed using software-radio front-ends as signal transmitters. This method allows to generate RF signals that replicate the case of two or more OFDM-based IEEE802.11 (a/g/p) frames being overlapping in time while being received at the antenna port of a real world receiver implementation.

Deliverable D.4.7 51 4.3. Part II - Resource Utilization Awareness in IEEE802.11 4. BTC

Using this methodology we were able to observe that the device-under-test receiver implementation is able to decode IEEE802.11 OFDM frames that are overlapping in time as soon as the difference in average received signal strength exceeds a certain threshold – a feature is known as physical layer capture. After adapting the interference model to reﬂect the measurements we were also able to quantify the impact on system-level simulation results. Additional to the observation of the capture effect, an important interference model assumption must be taken into question after analyzing the measurement results: treating interfering OFDM signals equivalent to AWGN noise contributions may lead to an overly pessimistic result in interference-limited scenarios. This insight is crucial for accurate estimation of individual transmission costs in the following, second part.

4.3. Part II - Resource Utilization Awareness in IEEE802.11

Whenever individual links in a wireless mesh network are not fully separated in either spatial or frequency domain, the algorithm that governs how the common radio resource is allocated amongst active nodes determines overall mesh performance. For an optimal decision in this allocation to take place, certain statistics about the physical layer need to made be accessible to routing algorithms, MAC schedulers, and last-but-not least to mesh network maintainers/users (in case of manual intervention). Unfortunately, within the majority of embedded wireless software stacks many vital lower-layer statistics are not exported to upper layers yet. Our goal is to close this gap. On the following pages we present some of the modifications to the linux kernel driver code and selected user-space tools we have implemented and are continuously testing in an 8-node indoor mesh. A first trial was started at this years BattleMesh in Maribor, Slowenia. This initial test was not successful - but many important lessons on kernel-level implementation issues have been learned during that week. Once integrated into the default software stacks (e.g. also included in OpenWRT releases), these modifications will allow to continuously and effortlessly keep track of the instantaneous channel load and spectral efficiency statistics for every link in the network.

4.3.1. Introduction

The performance of any wireless mesh that is not comprised of fully decoupled point-to-point radio links critically depends on how efficiently the finite signal space is shared amongst network nodes. Adaptive algorithms within the network stack need to be able to maximize the benefit/cost ratio over all transmission events in order to preserve common radio resources [33]. When accurate statistical information for this optimization is not available as a metric at the input of scheduling and routing algorithms the overall network performance deteriorates very fast with increasing node density and traffic demand. Improvement through manual intervention (e.g. channel re-assignment or re-location of nodes) is also challenging when informative statistical data is not available. Interestingly and somewhat surprisingly, most link metrics that are actually used in todays meshes lack a feature that seems crucial to optimally solve the common resource sharing problem at hand: they are not based on direct observations at link-layer. In most cases they are based on indirect observations (via active probing) at network layer. These estimation methods actively consume valuable radio resources themselves and have very limited accuracy when compared to direct observation at the physical layer. This deficiency is likely caused by the fact that access to most relevant link-layer statistics was previously hard if not impossible to implement on typical wireless mesh software platforms. Network interface drivers did not expose them at all or in a proprietary and therefore limited

Deliverable D.4.7 52 4. BTC 4.3. Part II - Resource Utilization Awareness in IEEE802.11 manner. However, these restrictions have been partly lifted by recent changes in the Linux wireless stack (using the SoftMAC architecture) and subsequently, new APIs have become available that allow to expose these statistics. Furthermore, appropriate cross-layer communication protocols have being drafted by the IEEE and the IETF that can be used to propagate those statistics in a non-proprietary manner [34]. In this document we discuss two essential real-time link-layer statistics and how to access and communicate them between layers on the currently most popular software platform in the wireless mesh developer community. We demonstrate how to extract high-resolution channel load and per-link spectral efﬁciency statistics using widespread Qualcomm/Atheros (ath9k) chipsets on an OpenWRT system. The necessary modiﬁcations to the linux kernel driver code and user-space tools do not interfere with the current network architecture and do not incur any considerable computational complexity or require additional hardware elements.

4.3.2. Related Work

Most adaptive routing protocols used in operational CWNs today are based on derivatives of the expected transmission count (ETX) metric [35]. ETX metrics are generated by measuring the loss rates at network layer of probing packets that are periodically broadcasted between neighboring nodes by the routing protocol (e.g. HELLO packets in OLSR). As already mentioned in the introduction, this indirect estimation has a number of negative effects and a plethora of alternative metrics to ETX have already been proposed in academia to mitigate this problem; ETT, WCETT, LAETT, EETT, MIC, iAWARE and many more [36]. However, very few of them have found their way into software stacks of todays CWNs as the respective link-layer functions have been kept closed in proprietary parts of the code tree [37] and no standardized cross-layer communication method was available. In other words, todays link metrics still fall short of giving insight into detailed characteristics of the link-layer itself, such as: currently perceived channel load, modulation and coding scheme (MCS) statistics and link-layer retransmission probabilities, just to name a few.

4.3.3. Channel Load Measurement

Within IEEE802.11 networks the so called Distributed Coordination Function (DCF) is used to or- chestrate medium access in a distributed manner by implementing a multiple-access with collision- avoidance (CSMA/CA) principle. In all IEEE 802.11 OFDM-based receive chains, channel load is deﬁned as the percentage of time that the instantaneous received power at the antenna input exceeds a constant known as Clear Channel Assessment (CCA) threshold [38]. This metric includes all sources that “block“ the channel in the respective part of the spectrum, such as: • IEEE 802.11 frame transmissions from distant nodes (received power above CCA threshold but lower than successful reception thresholds). • Interference power spillover from nearby IEEE 802.11 transmissions on adjacent channels. • Energy emitted from nearby non-IEEE 802.11 sources (inculding intermodulation products from high-power cellular towers).

One way to extract channel load ﬁgures is to export internal counters of the clear channel assessment (CCA) function within the DCF. Information gathered from posts on the ath9k mailing list [39] indicated that these statistics can be access within current Qaulcomm/Atheros ath9k-based chipsets by

Deliverable D.4.7 53 4.3. Part II - Resource Utilization Awareness in IEEE802.11 4. BTC exporting the MAC state cycle counters at kernel-driver level. Those counters can then be used to quantify the time intervals the radio front-end is in receiving, transmit or idle state with millisecond accuracy. This kind of insight comes without any measurement overhead and cannot possibly be gained by network layer probing. In order to demonstrate the potential of this method Figure 4.10 shows a short test-run of busy-cycle counter extraction on 8 co-located nodes that are tuned to adjacent channels each (channels 3 ... 10). At second 40 a short downstream session on a distant client-AP on channel 3 is started and lasts until second 60. This shows the expected behavior and corroborates the effectiveness of channel load tracking: power injected on one channel causes adjacent channels being sensed as busy. Once link metrics incorporate channel busy counters they can fully convey the interdependency that exists between links that cannot be fully orthogonalized in signal space (by operating frequency, geographical distance and/or directional antenna selectivity).

0.5 Channel Busy Ratio 0 3 4 5 6 7 120 8 100 Channel Number 80 9 60 10 40 20 0 Time [seconds] Figure 4.10: This ﬁgure shows the channel busy ratio readout at 8 co-located nodes that are operating on channels 3 to 10. At time 40 a nearby AP-client pair starts to transmitter on channel 3. The spillover to adjacent channels (up to channel 8) is clearly visible.

The relevance of busy cycle counters is further exempliﬁed by the fact that they are already used in an automatic channel selection (ACS) [40] algorithm in very recent versions of hostapd (a Linux user-space tool that implements management functions in infrastructure mode).

4.3.4. Channel Load Broadcasting

During our literature research on channel load statistics we discovered that the IEEE 802.11 standard documents deﬁne a metric called QOS-enhanced Basic Service Set (QBSS) Load. The following deﬁnition in section 8.4.2.30 of [1] indicates that this is indeed equivalent to the channel load we describe above:

[...] the Channel Utilization ﬁeld is deﬁned as the percentage of time, linearly scaled with 255 representing 100%, that the AP sensed the medium was busy [...]

This metric is meant to be communicated at regular intervals from access points to clients within the payload of IEEE 802.11 beacon frames [41]. But its applicability is not limited to infrastructure mode. In general, QBSS Load Elements offer are standardized method to continuously communicate perceived channel load data to all 1-hop neighbors with the wireless network.

Deliverable D.4.7 54 4. BTC 4.3. Part II - Resource Utilization Awareness in IEEE802.11

The original motivation of this function (introduced in the IEEE 802.11k amendment) was to improve QoS of low-latency services in 802.11 access networks. Handover decisions based on the channel congestion information in QBSS beacon elements (instead of signal strength) have a much higher chance of being successful and not causing a dropout during an interactive voice/video session.

Figure 4.11: Current versions of Wireshark support parsing of the QBSS beacon element as deﬁned in [1], allowing us to validate our current implementation.

However, to the best of our knowledge, QBSS tag transmission within IEEE 802.11 beacons has not yet been implemented in any open source software stack. The kernel driver code trees and mac80211 subsystems of the current Linux wireless stack do not contain the necessary routines. We therefore prototypically implemented this feature on a Qualcomm/Atheros ath9k-based Open- WRT system (latest release at the time: 14.07 barrier breaker). The modiﬁcations included regular polling of busy cycle counters from the ath9k kernel driver APIs, averaging and appending them to the list of default beacon tags in the transmit path of the mac80211 subsystem.

4.3.5. Modiﬁcations to the mac80211 Subsystem

The original patch implemented all necessary functions in the ath9k driver code structure instead of mac80211 sublayer structure. We collected very valuable feedback on these design decisions during the first deployment test at the 8th WirelessBattleMesh (WBMv8) in Maribor, Slowenia, August 2015. Figure 4.12 shows the most important part of the patch in its current, revised form. Since the beacon element update function is called by a software-interrupt handler (beacon tasklet) special care has to be taken when selecting the appropriate process synchronization (locking) method. All functions have been moved to the mac80211 sublayer, which makes the modification effective for all wireless chipsets/drivers that implement the get survey function: p54, b43legacy, mwl8k, b43, rt2800, ath9k, ath10k, carl9170, ath5k, wl1251, libertas tf, mvm (part of iwlwifi).

4.3.6. Channel Load in HORST

Figure 4.11 shows the final result: the structure of the final QBSS Load Element tag as seen at the output of a recent version of Wireshark. We also patched the open source network scanner tool horst [42] to parse and display the information contained in the newly added QBSS tags. The bars to the right of Figure 4.13 show the perceived local channel loads as they are now broadcasted in beacons. All stations are operating on the same channel but are spatially distributed within our premises. One can see that in this case the perceived instantaneous channel occupancy differs because of the spatial de-coupling between node locations. The additional overhead that is caused by transmitting additional 7 bytes per beacon interval is low compared to typical link speeds in IEEE 802.11n. Compared to this the benefits of broadcasting

Deliverable D.4.7 55 4.3. Part II - Resource Utilization Awareness in IEEE802.11 4. BTC

Figure 4.12: Current implementation of the beacon add qbss function. The block at line 16-41 needs to be further improved by reducing its complexity and replacing the rcu read lock – we are not sure if it is save to use that method in the given context on all processing platforms. channel load via regular beacons can be enormous: they reveal the local radio resource availability, something that cannot be inferred by signal-strength based views. Also, by evaluating the correlation over time between perceived channel loads from neighboring nodes on the same channel the spatial coupling between links can be estimated - previously hidden node relations can be easily detected.

4.3.7. Relation to License Assisted Access

The cellular industry is currently very actively pushing standardization and regulatory bodies to pave the way for operators being able to opportunistically use unlicensed bands via integration of a functionality referred to as License Assisted Access (LAA) into ﬁxed-line operator controlled home gateways and public mobile network small cells. This scheme was previously known as LTE-U and can be

Deliverable D.4.7 56 4. BTC 4.3. Part II - Resource Utilization Awareness in IEEE802.11

Figure 4.13: Output of a patched version of the network scanning tool called ’horst’. It shows channel load broadcasted from appropriately modified nodes operating on the same channel but located in different rooms of our office building, giving insight into the global radio resource situation and the spatial isolation between nodes. seen as a slightly modified version of LTE that shall be able to co-exist with IEEE 802.11 users in the 5.2-5.8GHz band. To some extend, LAA would put more control over unlicensed radio resources in operators hands. Early studies indicate that a widespread rollout of LAA would have significant negative impact on the performance of any co-located IEEE 802.11 based mesh network. In such cases, channel load statistics would allow to pinpoint that IEEE 802.11 mesh performance deterioration is in fact due to a nearby active LAA small cell and not due to self interference.

4.3.8. Spectral Efﬁciency Measurement

We have shown that busy-cycle counters can be used to provide valuable insight into the instantaneous radio resource availability situation and interdependencies between individual links but they do not convey how efficient the channel resources are spent whenever a transmission takes place. Therefore, links which would support efficient channel code rates cannot be identified and preferred over airtime- consuming low-SNIR links. Motivated by this we have started to investigate a passive measurement method that aims to provide the necessary input data for evaluating link efficiency in an information-theoretic sense. The goal is to be able to assign an effective link-layer throughput (in bits/second) to each destination neighbor node as soon as unicast transmissions to this MAC address take place. Figure 4.14 shows the modified internal data structures of Minstrel-HT (the link layer modulation and coding scheme adaption algorithm in ath9k device drivers). Those statistics now allow to calculate air-time consumption and the respective effective bit-rate of unicast transmissions per MAC address. Once conveyed to upper layers (e.g. using DLEP [34]), algorithms using maximum common resource utilization efficiency criteria for every forwarding and scheduling decision become feasible. Figure 4.15 depicts an exemplary illustration of the concept. Unicast transmissions that contribute to the channel occupancy from the viewpoint of the transmitter are compared by their effective bit-rate that has been achieved (including link-layer retransmissions). In this case unicast transmissions to three neighbor nodes do individually render certain areas of the total available downstream channel time unusable for more efficient transmission activity by any other source that the local signal space needs to be shared with. The goal is to arrive at a link metric that describes the relation between the attainable spectrum efficiency and the percentage of channel resources (airtime) that each link actually consumes while active. This would significantly decrease the chances that network throughput quickly deflates under traffic demands within the mesh.

Deliverable D.4.7 57 4.3. Part II - Resource Utilization Awareness in IEEE802.11 4. BTC

Figure 4.14: Modiﬁed Minstrel-HT modulation and coding scheme statistics, those tables now also contain the number of successfully/not-successfully transmitted bytes to each unicast address.

Figure 4.15: Illustration of the concept of a resource-utilization-map in transmit direction. The colored areas correspond to the consumption of radio resources that do not lead to a beneﬁt in terms of bits being successfully propagated through the network.

4.3.9. Summary of Part II

Currently, from the viewpoint of higher layers, the occurrence of interference in IEEE802.11 networks is invisible. Active probing at the network layer cannot provide the necessary insight into channel load and spectral efficiency within IEEE 802.11 links and thus fails to provide necessary statistics to accurately describe ”how much of the common radio resources are consumed how efficiently by which nodes”. Methods to extract and communicate the appropriate internal datasets across layers and nodes in non- proprietary manner have become feasible recently and this report documents the preliminary results of our implementation efforts. The channel-load and spectral efficiency estimation methods described in this paper are continuously being tested and further validated using the 10-node OpenWRT/TP-WDR4300/ath9k indoor mesh that

Deliverable D.4.7 58 4. BTC 4.4. Concluding Remarks has been setup speciﬁcally for that purpose. Once all quality checks have been concluded the patches will be sent to main mac80211 subsystem repositories (wireless-testing, as of 2015 maintained by John W. Linville).

4.4. Concluding Remarks

The idea of resource consumption awareness is inline with the lessons that can be learnt from human history. Tragedy of the commons occur in cases when appropriate statistics are not communicated, causing individual participants to not being able to understand their contributions (and those of neighboring peers) relative to total resource availability. Societies are not able to optimize global resource efﬁciency (in a continuous, self-regulative manner) in the absence of such statistics. Therefore, in interference-limited wireless mesh networks a function to measure and communicate individual radio resource utilization efﬁciency is needed. Without this, local agents (humans and routing protocol daemons alike) cannot maintain high network performance in a distributed fashion – no matter which algorithm is used – the necessary input data for that kind of optimization is missing. During this project we tried to take the necessary next steps into removing these shortcomings.

Deliverable D.4.7 59 Part II.

Social experiments

60 5. Citizen Square Kilometer

5.1. Introduction

The main aim of CitizenSqKm is to benefit the neighbourhood by engaging a community network, made up of students, local administrations and local entities, in the development of their own community using geolocation technologies, based on a commons based computer network (guifi.net). The local community network, promoted by the project core researchers, actively participates through pedagogical and ludic (play) experiences, in the discovery and improvement of the neighbourhood by collecting and classifying data related to it, by creating geolocated layers of relevant information and itineraries. A map has been created for people to make an inventory of the “things” in the neighbourhood; like institutions, commerces, historic buildings, plants, temperatures, etc.; with the aim to overcome and better manage the information linked to the territory and classified by author, source and topic. Par- ticipants conduct a census of the land, its inhabitants, its infrastructures, its services, its history and its nature and as they do so they take full advantage of data already collected by the inhabitants and open data coming from public administrations and sensors. This process of ‘civic reappropriation of data’ engages citizens and local administrations in the development of their own community. A methodology has been developed to observe and measure the interactions among individuals, through Social Network Analysis (SNA) and also their interests, those of the individuals and those of the organisations, applying Content Analysis Techniques, so it can later identify change, if change occurs. The goal for this experimentation is to contribute to an increase in public awareness about the potential of collecting data, local content and sharing knowledge, and to build and strengthen the community network of the selected quarter thanks to the deployment of a commons based computer network infrastructure and training and empowering citizens to work collaboratively in activities that improve the neighbourhood. And to make CitizenSqKm a model to be reproduced and applied to other communities worldwide.

5.2. Background

Due to its ubiquitous attributes, digitized media blur the boundaries between the roles of public administration, journalism, entertainment or educational sectors. Anyone who has a mobile device connected to the Internet can create, collect, process and share data massively, geotagged and in real time. All existing media can be translated into numerical data accessible for computers (Lev Manovich, 2001). For citizens to be engaged and informed it is not sufﬁcient, anymore, to only conduct a simple description and transmission of facts to audiences; the communication process needs to instigate citizens participation, and it needs to be able to inﬂuence and help citizens to exchange ideas to create some form of ‘democratic debate’.

61 5.2. Background 5. Citizensqkm

The goal of this CitizenSqKm’s experimentation was to, by using educational methodologies, convert the selected district in a model to explore how a community reacts to a new platform where digitized information is created, collected, guarded, processed and disseminated by citizens (students, researchers, volunteers, possibly public officers, entrepreneurs or journalists) in a collective and structured effort. It followed a model of social innovation based on collaborative production by citizens through location based technologies connected to the Internet. CitizenSqKm used the Confine concept of Commons (network infrastructure shared and deployed by citizens in benefit of the whole district citizen network), and applied this same concept to location based content and data created and shared by inhabitants in the selected area. CitizenSqKm starting point was on one side, the coordination of a research project conducted by Itinerarium at two secondary schools: Escola Joan XXIII, Bellvitge, Barcelona, and Padre Piquer in Madrid under Fundacion´ Telefonica´ leadership, demonstrating how service learning combined with mobile location based activities can enhance motivation and improve academic results of students at risk of social exclusion. On the other side, the development of ‘Eduloc’ (http://www.eduloc.net/en); a Geolocation technological platform for a school based learning project and a tool for young students to learn through service learning methods, located narratives and gaming. ‘Eduloc’ is an active platform that uses mobile devices for teachers, students and families to create itineraries, scenarios and experiences based on location. CitizenSqKm built on these two previous experiences and took a step further to involve not only students but also civil society, the community network of the selected district, in a new commons based computer network. The pilot has been implemented in Barcelona, where access to the internet (and especially to download or upload media files) is not always available in open spaces. It created a geolocation platform open to the citizenship to develop a social educational experiment (based on Itinerarium methodology: Adopt, Adapt, Create and Share) to observe and conduct an experimentally-driven research on the community network following Ethnographic Active Research, to demonstrate the potential of service learning projects based on location, their possibilities to provide a rich and inclusive learning environment for all citizens and to encourage entrepreneurship involving them, actively, in their community, and also offering means of participation to educational networks and local civic institutions, such as institutions for the elders, the youth, migrants, disabled, women, children,.. the development of the arts, new technologies, among others. CitizenSqKm identifies a square kilometer area surrounding the school and invites local entities and individuals to create points of interest and itineraries on the platform. When someone strolls through the area with a GPS mobile device, these routes are activated and can be explored. Participants will be encouraged to not only publish information in different mediums (text, audio, video, images) but also identify elements of the neighbourhood which need to be fixed or improved, and design emotional experiences for visitors by introducing game elements. This CitizenSqKm is a longitudinal research study of the designing, developing, implementing and uptake of the platform, and it explores to what degree geomedia tools can be active and inclusive, and the qualities and limitations of education and communication when we use these tools. Experiments in Poblenou drew on both, Itinerarium’s project learning educational methodologies and the Ethnographic Action Research (EAR) methodology, designed for Information and Commu- nication Technology (ICT) initiatives, that combines research with project development, in a multidisciplinary study. The guiding research question is: “How does CitizenSqKm allow for new and contemporary notions of networked civic engagement?” Ethnographic and participatory techniques were used not only to guide the research process and action research but also to link the research back to the CitizenSqKm project through the development and

Deliverable D.4.7 62 5. Citizensqkm 5.3. Test Setup and Results planning of activities. The main objectives of CitizenSqKm project were to convert the selected district in a model of how crowd sourced contribution of active citizenship using location based technologies can engage citizens and local administrations in community network developments and enrich the quality of the experience of participants in community networks. To create an open and scalable model, with tools developed for the community, to beneﬁt from the data created by citizens and an opportunity for external researchers to learn more about community networks. It was expected that having a better and deeper knowledge about the local areas would place students in a privileged position, to preserve public property, to improve living conditions, and to make the area a better place. Different Service Learning projects that arose within the project have already had a positive impact in the district, because their main goal was to improve quality of life of the community and produce change wherever possible. The visibility of service learning projects accomplished helped, in its turn, to increase public awareness about the potential of the community network.

5.3. Test Setup and Results

5.3.1. An extension of the guiﬁ.net community network

Several guiﬁ.net antennas have been installed: the main one at Itinerarium headquarters, which connects to guiﬁ.net and expands the Catalan community network through a QmP antenna, which gives connectivity to several neighbours: Universitat Pompeu Fabra, Escola Vila Ol´ımpica and Cooperativa Flor de Maig.

Figure 5.1: Active nodes map

A new antenna - node- was installed and connected on the top of the building where Itinerarium has its ofﬁces1. Installed items are: • MikroTik UPA ( acts as router) • NanoBeam400 (links to another identical antenna at Almogavers) Sectorial 120 with rocket M5. (network distribution)

1http://guifi.net/ca/node/74188/: (DrTrueta183)

Deliverable D.4.7 63 5.3. Test Setup and Results 5. Citizensqkm

• GuiﬁLab (20th Nov) A second node was also deployed at a local Secondary School (Institut Poblenou) by Routek, a company focusing on the development of telecommunication projects and the research of their impact in the society. CitizenSqKm also provided the hardware components and the know-how (with several workshops and guides) for new nodes to be deployed around the area. Local organisations and citizens in the selected district were encouraged to use it.

Figure 5.2: Interactions Map

CitizenSqKm platform was replicated within a guifi.net server, the community network can now offer CitizenSqKm as a service to the community, within the Clommunity Project http://clommunity- project.eu/ A server with Cloudy distro was installed in the UPC and was connected to the guifi.net network, with the software developed in CitizenSqKm project. Also a mirror server was installed in the Hangar guifi.net node in Barcelona. CitzenSqKm platform is now accessible via Internet (162.255.119.254) or via guifi.net network (10.139.40.64 and 10.139.94.109, or using a guifi.net DNS server with the same domain).

5.3.2. Itinerarium’s geolocation platform

Itinerarium’s geolocation platform was customized for the project and it also became one more element within a wider communicative ecology with 5 layers of communication: Forum, Geolocation Platform, Social Media, Survey and Community Network access. Layer 1. The Forum became a space to exchange information and views among participants, to enable social interaction between users, public exposure for the projects and a place to share common resources to strengthen the network community links. A work in progress blog was created in Word- press (Guide 1). http://blog.citizensqkm.net/, and private Field Notes and emails exchanged were kept. A blog with continuous information about how the project evolves (work in progress) was kept.

Deliverable D.4.7 64 5. Citizensqkm 5.3. Test Setup and Results

Layer 2. The Geolocation Platform (with a web page, and mobile applications for iOS and Android and connected to guifi.net nodes or to the Internet) was created, for data to be copied and shared, infinitely, massively, geotagged and in real time. Participants were able to create and access content and data produced by themselves, collected by sensors, or published by public institutions (open data) at the online platform and via mobile applications, from the Secondary School, their own homes, or a mobile device. Also for the sake of the experiment personal data were collected, kept and anonymised. Itinerarium’s geolocation platform has been customized to enable the exportation of open data, by using data files in KML format, but it needs further development for it to work properly. Layer 3. Social Media accounts were created and curated. Layer 4. Survey platform was used to conduct semi-structured online surveys. Layer 5. Logs between the platform and the community or internet network, measuring data, and interactions among actors and with contents were also established in the geolocation platform.

5.3.3. Ethnography and participatory techniques to Adopt, Adapt, Create and Share

CitizenSqKm uses service and project learning as incentive models, with Itinerarium methodology, to encourage users to participate in the community network and at the same time CitizenSqKm becomes a platform at the service of the community network and at the service of the EAR research. Itinerarium methodology integrates key elements of EAR methodology, for the initiative to be able to be adjusted periodically recognising and responding to local social, political, cultural and economic contexts and to help researchers share, store, manage and analyse data. Researchers will engage the participants, to include their ways of making sense of the world and themselves in their evaluations of projects, following the motto: Adopt, Adapt, Create and Share. Researchers (social-cultural anima- tors) provided training (talks and workshops) for participants to learn how to take part in CitizenSqKm applying Itinerarium methodology: Adopt, Adapt, Create and Share and to integrate ethnography and participatory techniques. Having spoken to the community (local associations, shop owners, foundations, NGO’s,...schools) and to other projects not related to this community, some participants became active on the platform and others gave advice or participated in some other way. As the process of creation happened, researchers collected data, reflecting on what they observed and recorded it in the form of field notes, conducting in-depth interviews, short questionnaire-based surveys, and using other tools to get feedback. The interests and needs of each organisation were recorded, and common interests were actively searched, looking at identifying serendipities. When core researchers did a first approach with the local organisations, they engaged public officers, entrepreneurs, volunteers, researchers, journalists, students, and other citizens (also artists, journalists, activists, cultural workers). For each contacted institution that wanted to support or participate in the project, a Field Note was opened, with the name of the organisation, type of institution; role, email address and name of each member. Later on, the date, the type and the content of each exchange: phone call, one-on-one, face-to-face, online meetings, workshops, note by the researcher, other (which includes blog post) were also registered. Emails exchanged between participants and core researchers were also kept. To create the paths of participation there were about 100 institutions directly contacted. Field Notes were filled for 49 of them, conversations held with 140 individuals and notes recorded . A total of 29 interlocutors worked in a technological organisation, 9 in a media outlet, 23 in primary schools, 13 in public organisations, 19 in civil organisations, 11 in high schools, 24 in research institutes, 6 in a university. The geolocation platform had the active participation of 450 users, most of which were students or teachers in schools and high schools. There were 5 Primary Schools, involved: Escola

Deliverable D.4.7 65 5.3. Test Setup and Results 5. Citizensqkm

Figure 5.3: Field notes template

Grevol,´ Escola Les Acacies,` Centre d’Estudis Montseny, Escola Vila Olimpica, Escola Voramar and 4 Secondary Schools: IES Salvador Espriu, IES Front Maritim, IES Poblenou, Institut Barri Besos.´ A survey to actors registered on the platform and to 270 new contacts, mainly institutions in Sant Mart´ı District. In this ﬁrst wave 94 people (34%) answered the questionnaire.

Figure 5.4: Communicative Ecology of Km2 Poblenou (May 2015)

This community worked together on ﬁnding existing projects and common interests among them. The resulting projects and scenarios created can be explored on CitizenSqKm’s blog: http://blog.citizensqkm.net/category/escenaris/ Two types of collaborative working processes were developed: • Collaborating with organisations or institutions to design a path of participation, working together CitizenSqKm and the collaborating institution, planning to use the geolocation platform with a speciﬁc aim. • Bringing together several projects: where the path of participation was designed by promoting the use of external projects, not related to CitizenSqKm, and geolocating the information collected by them. Once the different collaborating initiatives were developed and tested, a guide

Deliverable D.4.7 66 5. Citizensqkm 5.3. Test Setup and Results

for its replication elsewhere, was produced and published.

5.3.3.1. Collaborating with organisations or institutions

The first step, was for the community to learn to use the technology to regain control over data. Workshops and guides were created, with inputs from guifi.net, the Open Technology Institute of the New America Foundation and with local high schools and primary schools, about how to use the geolocation platform, how community networks function and the potential of collecting data with sensors. Tutorial to use CitizenSqKm Geolocation Platform and for the project location based main activities to be developed following the Adopt, Adapt, Create and Share methodology. Students, local civic institutions, and other citizens (elders) were encouraged to participate using the available tutorials on how to use the platform or being part of the workshops prepared for citizens and students to learn to use the platform2. Workshops to Expanding the network were organised, with practical workshop on the extension of a community network (guifi.net) for participants to learn to install an antenna or a router, configure it, and connect it to the guifi.net network. Guides were prepared with Commotion Wireless a free, open-source communication tool that uses wireless devices to create decentralized mesh networks. Documentation to get started was translated into Catalan and contributed to the Open Technology Institute of the New America Foundation. Guides are published for anyone to access, download, copy and share. (Guide 2) Together with members of guifi.net in Barcelona, the use of the community network among the community of El Poblenou, was promoted. Workshops for citizens were conducted to learn how the community network functions, to install nodes and to conduct a few experiments collecting data3.A protocol to replicate this workshop Citizen guide to expand Community Networks GUIFI.NET / qMp has been prepared and is also shared. (Guide 3) Citizen guide to create a portable weather station To learn how to create networks of sensors and connect them to guifi.net. This allows citizens to build small weather stations, creating indexes of air pollution, noise pollution, air humidity, etc. and to identify and compare microclimates4. Journalistic initiative to geolocate journalistic, current affairs and controversial issues about the neighbourhood, such as its transformation, the growth of tourism, the lack of services, also emerged. Con- versations with local and hyper-local media helped create the different layers and categories in the platform and suggested to create a space to keep a calendar, and to have the possibility to merge scenarios with information uploaded by different media outlets, to be able to compare.

5.3.3.2. Bringing together several projects

Urban Flora and Allergy Together with scientists at the Aerobiological Information Point (Autonomous University of Barcelona), a path to geolocate and to conduct a phenology of plants in one speciﬁc area, was created. PIA reports on pollen concentration in Barcelona, according to one single sensor located in the city, help from citizen observation is invaluable to them, and as a consequence, CitizenSqKm

2http://blog.citizensqkm.net/tutorials/ 3http://blog.citizensqkm.net/dades-en-xarxa/ 4Citizen guide at Google Docs

Deliverable D.4.7 67 5.3. Test Setup and Results 5. Citizensqkm core researchers and PIA experts designed the method to collect the missing information to predict the amount of pollen in the streets using CitizenSqKm geolocation platform. A Urban Flora and Allergy Citizen Guide was created and is available. (Guide 5)

Biodiversity Experts say that biodiversity in Barcelona shouldn’t be observed only to avoid pollen, but we have a very rich urban biodiversity and the majority of animals found in the city are and wildlife are protected. Several schools and high schools in the neighbourhood were already doing school outings to explore the biodiversity near the city, and suggested to instead explore the inner city urban biodiversity by geolocating it. A Citizen guide to create a Local Census of Urban Biodiversity was created and is available. (Guide 5)

Historical Heritage For citizens and students to deepen their knowledge on the historical, artistic and cultural local heritage that surround them, collaboration with the local historical archives, public library, and private citizens initiatives was actively searched. The work of Xavier Badia (a former employee of one of the local high schools) emerged, and his graphic archive collected over 30 years and kept in the form of a PDF, was offered to be processed and geolocated. (Guide 5) CitizenSqKm actively searched the assessment of the local Amical Wikimedia to design a strategy to process the graphic archive following their advice before its geolocation, a Wikiproject was created5 and the advice incorporated in the Citizen guide to keep the local historical memory. (Guide 5)

The Optimal Path Core researchers spoke to the association of Fathers and Mothers of one school, and to organisations with projects not directly related to the local community of el Poblenou, such as CitiSense, Open Systems, PIA (Point of Aerobiological Information) and Wikimedia. After seeing how the platform works, each organisation indicated what they would use the platform for. The Association of Fathers and Mothers wanted to use it to geolocate the project “The way to school, friendly spaces” (Camins Escolars, espais amics), a public administration initiative where children going to school on their own can find safety in participating local shops. These shops are marked with an indentifying sticker on the window. The Point of Aerobiological Information, as explained above, needed help from citizen observation, to conduct phenologies of specific plants. Open Systems needed to evaluate their BeePath phone application to do experiments on human mobility. CitiSense, a project to develop “citizens’ observatories” wanted to test their sensors to measure air quality. Altogether, core researchers and members of the Association of Mothers and Fathers, PIA, BeePath and CitiSense planned for a project to involve citizens massively, without having to conduct 5 different projects, and the common project was to find which is the Optimal Path to go from home to school or to work and back. Perhaps the best route is the most fun, the safer or the healthier, not the fastest or the shortest one. A Citizen guide to find ’The Optimal Path’ was created and is available (Guide 5). These diverse paths of participation, and each element created by a participant, an individual or institution, were shared with the whole network to enrich the overall citizen experience. Citizen- SqKm project had continuous activities to achieve creating a true community network with different stakeholders: teachers, students, their families, local institutions and the administration. Projects, itineraries, scenarios started to be created and shared via social media.

5https://ca.wikipedia.org/wiki/Viquiprojecte:Fons_gr%C3%A0fic_Xavier_Badia

Deliverable D.4.7 68 5. Citizensqkm 5.3. Test Setup and Results

Figure 5.5: Sharing projects via social media

Figure 5.6: Social Media Analytics

As users designed their paths of participation, engaged their peers, and gathered and published the information on the Geolocation Platform, they were leaving a trail of data that was collected by each of the 5 layers, generating the Communicative Ecology Network Proﬁle. The Forum became a space to exchange information and views among participants. The Geolocation Platform successfully became a gallery to organise and systematize the knowledge that belongs to the community. Social Media accounts (Twitter) were successfully set to become a gallery to conduct a public conversation among the community. Surveys deepened in the opinions and views of the participants. The Community or Internet Network were also tested as channels through which data travels and is stored. Research and Participation Methodology The research project was based on experimentation, designing a comprehensive qualitative and quan-

Deliverable D.4.7 69 5.3. Test Setup and Results 5. Citizensqkm titative methodology of assessment based on multilayer analysis, iterative planning and on cyclical evaluation (Guide 5), with a methodology, both for participation and for research, to collect the trail of data left by users as they designed their paths of participation, engaged their peers, and gathered and published the information. Data being collected on 5 layers which form the Communicative Ecology Network Proﬁle (Forum, Geolocation Platform, Social Media, Survey, Community Network)

Figure 5.7: Communicative Ecology (Layers)

A database was created, as shown in ﬁgure 5.8.

Figure 5.8: Communicative Ecology Network Data (May 2015)

Information from Layer 1 is collected in Columns A to E which include the email of the participant, i.e.: ID indicators Table - EMAIL 1. Personal information, such as the ﬁrst part of the email address has been deleted for the sake of anonymity of the participants. A number has been allocated to each participant, i.e.: Participant number: email 1001). The role of the participant in its institution, the institution they are members of and the type of institution is also stated. Information from Layer 2 is collected in Columns F to I, with the role of each user in the platform, their names have been deleted, for anonymity sake, but their Id number on the platform has been kept. Information from Layer 2, is also collected to describe the type of user: Dynamising Agent; Professor / Coordinator; Student / Participant or Other. Information from Layer 3 (social media accounts) has also been deleted and substituted by a corresponding Id number (correlated to the ID Indicator, i.e.: Participant number: twitter 33) in column J. Information from Layer 4 (survey) has also been given a 0 in column K, if the survey has not been answered and a 1 if it has.

Deliverable D.4.7 70 5. Citizensqkm 5.4. Main Achievements and Challenges

Having collected these data will allow for a systematic analysis of the relationships developed among participants and of the content collected and generated on the platform for educational, scientiﬁc, journalistic (and any other discipline) projects. The outcome of such analysis will be used in turn, to inform the platform and the educational, scientiﬁc, journalistic projects of the community. Con- sequently, both research and participation in the project will simultaneously observe, measure and enrich social interaction within the community.

5.4. Main Achievements and Challenges

The participation and research methodology to designing a communicative ecology and methodology will allow to measure the transformative impact that CitizenSqKm may have on its community. To be able to replicate and escalate the project elsewhere and conduct it in new communities, tutorials, guides, forms and protocols have been created, and they are enclosed in this documentation pack. The difﬁculties encountered in the experience in Poblenou and the strategies developed to overcome them, can be useful to inform other future projects in new communities.

Outsiders Core researchers at CitizenSqKm are outsiders in all of the segments within the community, although the headquarters of the project were located within the neighbourhood when the experiment began, the researchers themselves are not members of any local organisation from the area, they are not part of the public administration, nor the school system, or the local community. Alliances have been constructed with members of the community, specially teachers and activists, but the time length of the project was too short to strengthen these alliances. The EAR methodology uses “media itself as a tool for action research: for exploring issues in a community as well as archiving, managing and collecting data and facilitating online networks of EAR researchers.” (Hearn et al. 2009). Local and hyper local media outlets got strongly involved with the project but the timings within their organisations were very slow and there wasn’t enough time to involve participating media outlets with other stakeholders from education and public administration, making the initiative truly multidisciplinary.

Bottom-up EAR and MSC methodologies were used in triangulation with more conventional techniques to enable participation and to promote a bottom-up flow of information. Despite this, many of the participants approached preferred ‘top-down’ forms of participation, such as direction from the the core researchers of how they wanted participants to contribute to the project, and what actions the participants were expected to undertake, when and for what duration. As a result, in the first stage a more active approach than initially planned was taken to structuring avenues for participation. CitizenSqKm was obviously breaking the common ecosystem. End-users (citizens, parents, scientific researchers,...) welcomed the project. Some middle-users (teachers, journalists, researchers, ...) also embraced it, a few didn’t like it so much. Most institutions (public administration, schools, organisations) couldn’t figure out how to deal with it, they felt it was disruptive. As such, it was observed that individuals, especially those working in institutions, are not used to receive open proposals, they expect to receive clear instructions on what is being asked from them and what is the time schedule. The negotiation, coordination, creation of the paths for participation was very long and costly, the first 6 months of the project were invested into creating the first projects to engage

Deliverable D.4.7 71 5.4. Main Achievements and Challenges 5. Citizensqkm participants. The schedule of the project had to be modiﬁed and all stages were postponed.

What community? CitizenSqKm Confine’s submission stated that its main aim was to “benefit the neighbourhood by engaging a community network, made up of students, local administrations and local entities, in the development of their own community using geolocation technologies, based on a commons based computer network (guifi.net)”. Despite the efforts to define the concept of “community” from the very start, it wasn’t clear to all that “community” referred to the wider community in the neighbourhood, the general community, citizens who do not know what a community network is, and not the “community network” itself. guifi.net initially suggested developing CitizenSqKm’s experiment in the Barcelona neighbourhood of El Raval, but soon suggested Poblenou would be a better community to work with for this project. The main reason being the annual event “Tallers Oberts Poblenou”, by which Poblenou Sense Fils had committed to provide connectivity to local ateliers6. Hangar7, a center for arts production and research, set up by the Association of Visual Arts of Catalonia (AAVC), was at the center of the publicly funded initative, but the project has not been delivered yet, and the direction of Hangar has not found the occasion to meet with CitizenSqKm’s team. Knowing the Catalan local network community Gufi.net was participating in the Confine project and having had first meetings with Confine participants and members of Guifnet; having chosen in conversations with guifi.net the specific neighbourhood in Barcelona where the experiment would be conducted, Poblenou; and having conducted a first meeting with Confine participants and the local guifi.net group, Poblenou Sense Fils; it was assumed by core researchers that the necessary communication channels had been established, and support to the experiment from Poblenou Sense Fils was a given. That assumption was wrong and the complexities of guifi.net’s community should have been taken into account.

No common goal CitizenSqKm researchers’ main goal was to find how to extend the community network in a way that the wider community could access it using mobile devices for the sake of the experiment. Guifi.net’s community’s main goal was to extend the community network’s number of nodes locally and strengthen infrastructure and mesh network, “at guifi.net we are agnostic regarding the use, we only want to create network”. Some in the community network considered CitizenSqKm to be “free riders”, not giving any benefits and not planning to give any benefits to the community network. The project’s budget had funds to buy parts for growing the community network, but it was very difficult to find agreement on first defining what local institutions would participate in the project, if they were schools a permit from public administration was needed, how the experiment would be conducted, to then deploy nodes for the community network taking into account the experiment and also the needs of the community network. Exchanges happened almost one to one, researchers approached one member of the community network and he would call another member of the community network to ask about CitizenSqKm project, how was it funded and who was behind it. They would have a discussion and get back to the core researchers. When group emails were sent out only few of the receivers would answer.

6http://www.poblenousenseﬁls.net/tallers-oberts-del-poblenou/ 7http://hangar.org/

Deliverable D.4.7 72 5. Citizensqkm 5.4. Main Achievements and Challenges

Finally, 7 months into the project, a member of guifi.net.net recommended researchers to register to the local guifi email list guifi[email protected]fi.net. That was very well received by members of gufi.net and increased enormously the level of collaboration in terms of extending the guifi.net community network locally and developing workshops for schools. Within the local guifi.net email list, several conversations flourished. The possibility of creating a “simple and popular node with very strict specifications to make it usable for students, shops, cafes, ... something easy and viral for urban environments”, was considered by some within the list, but the idea was not followed up. Another conversation emerged about what is the community within a community network. Some guifi.net members were worried the local community might be misled by CitizenSqKm and made believe mobile connectivity from the street would be available. This is something guifi.net does not want to do, however other community networks within Confine do offer this. If guifi.net.net’s mobile node -an open transmission station or wireless telecommunications infrastructure that can be used in the urban space and connected to other digital networks- had been available or a new one had been developed interactions could have been measured, and other observations could have been done. Also experiments with human mobility apps could have been conducted and sensors could have been connected to the mobile node as well.

No common problem to overcome, either The EAR approach requires a high level of commitment by organisations, and is easier to apply in developing communities, where there is a prevalence of high need. In communities that don’t require assistance, having a common problem to overcome may work to create the necessary level of commitment. In the neighbourhood of Poblenou, as elsewhere in Barcelona, Internet connectivity from mobile devices in the street is limited. The community also has problems adopting new technologies, having access to computers and mobile devices, and having access to Internet connectivity and signal strenght. If CitizenSqKm wants to test if it can help citizens feel they can create an impact on their environment, having a mobile device connected to other users, they will need connectivity, ideally provided by network communities. Researchers in CitizenSqKm assumed they were tackling, together with other members of Gufinet participating in Confine, the problem of poor access to mobile internet connection and use of mobile devices by the community. But as noted above, not all members of Gufinet participating in Con- fine shared that view. Neither the public administration institutions had an intention to contribute to overcome such issues. To increase connectivity and install an antenna on the top of a public building, and especially on a school building, permission from the local administration is required, and the timings for permission are too slow to manage to install a significant number of antennas over the duration of the experiment. Core researchers in CitizenSqKm also searched for synergies for the community to have access to mobile devices, during the experiment and beyond. Cibernarium,` the programme for IT skills acqui- sition and diffusion of Barcelona Activa (Barcelona City Council) did have mobile devices for the community but the devices must be used inside the building; the public library could have lend 1 or 2 devices for a specific activity; and the education department could also lend about 10 devices for a specific activity, but not regularly. CitizenSqKm’s wider community (in Poblenou) seems to not have one single uniting problem to fight against, but many small problems that affect sub-communities within, such as: difficulty in

Deliverable D.4.7 73 5.5. Conclusions 5. Citizensqkm communicating to the wider community, issues in education, not knowing the community, too many tourists dilute the community (neighbours don’t know each other), lots of disconnected institutions and projects that overlap but ignore each other.

Lack of co-planning and collaborating skills As noted above, a lack of co-planning and collaborating skills, especially within institutions, was identiﬁed. Encouraged by several institutions (Cibernarium, Fablab, Public Library) which expressed their interest in being more involved with the local community, core researchers tried to develop intergenerational projects across institutions, but didn’t succeed.

Mapping walks through the area of Sant Mart´ı using images from the historical archive of Poblenou, and promoting work in teams of students and grandparents from Casal d’avis de Can Saladrigues where, with the support of the City Council and ﬁnancing from La Caixa (local bank), UPC university is training elders in using new technologies. Mapping shops participating in the City Council’s initiative ‘Camins Escolars’ (the way to school, by IMEB) with the support of the Barcelona Activa (Barcelona City Council) training of shop owners through Cibernarium. Mapping local services for older people, an initiative suggested in the very beginning of the project by a local publicly funded institution (Apropem-nos) which at the end of the 12 months of the duration of the experiment and after numerous meetings hadn’t produced the list of the services to be mapped. Where collaboration action has occurred, great success has been recorded, and often it has been thanks to crucial individuals and organisations, such as scientists in research organisations, mothers and fathers as well as speciﬁc teachers at schools, activists in local associations, and some journalists.

The challenge of time The experiment has been successful in setting a model to be replicated in the future. CitizenSqKm is at the right place for its second iteration, to test, replicate and scale it. The platform had to be adapted for the project. The research had set out to adapt the platform to the needs of the users, and to do that it was necessary to approach a new community and possible users and explain the potential of the platform without having a properly functioning platform. It has been difﬁcult but the experiment has managed to effectively adapt the platform taking into account the views of the users in an adaptive planning approach and continuous improvement of the platform. It has succeeded in sufﬁciently engaging the community and their ongoing projects, for the platform to continue to be used in the future by this community without needing further encouragement. The necessary guides to replicate the project and all materials produced are open and available for everyone to contribute to them replicate the experiment with a tested methodology.

5.5. Conclusions

CitizenSqKm is a humble contribution to improve European social platform services, educational models, and to enhance the development of Information and Communication Technologies for local communities. The project contributes to a better governance and policy making in new social practices of innovation; it is already being replicated locally, in three cities in the province of Barcelona; and a possible pilot to be conducted in South Africa is being explored. CitizenSqKm has been a truly multidisciplinary project, combining research teams from public policies, Internet science, ethnography, management studies, law, pedagogy, engineering and computer science, each of them

Deliverable D.4.7 74 5. Citizensqkm 5.5. Conclusions already involved in an interdisciplinary ﬁeld or frontier research, and developing non-traditional approaches. From a methodological innovation perspective, CitizenSqKm has been an applied case of social innovation based on geolocated data, adding value to the state of the art in the social science. The analysis has been inspired by the pioneering approach, known as Ethnographic Action Research (EAR), speciﬁcally designed for Information and Communication Technology (ICT) initiatives. EAR’s methodology combines research with project development, which has been guided by IGOPnet’s expertise and Itinerarium’s pedagogical and learning located methodology, tested internationally and renowned for its potential to increase students’ motivation and academic results. The methodology followed was described on a Paper published at the 8th IDIA Conference, the International Development Informatics Association. ICTs for inclusive communities in developing societies, “An experimental methodology to promote and evaluate the use of community networks for civic engagement”.

Local community of selected district Encouraged cooperation between different sectors of the population, such as school students and local entities. CitizenSqKm enhances the capability to connect distributed knowledge, skills and competencies, and promotes informational and social inclusion. The project provided the tools and technologies for the local community to engage and collaborate, increasing the degree and quality of the knowledge citizens have about the area of the selected district, helping identify and solve common problems. The project has generated a citizen driven community of social innovation, with a strong potential to create not only entrepreneurial attitudes but also employment opportunities among the local youth. This community of innovation will have continuity over time, lead by participating research institutes, neighborhood associations, media outlets and schools; their members, workers, teachers, students, and their families who will be able to follow and modify the guides beyond the development of this project, creating useful and visible impact for the citizens of the selected area.

Deliverable D.4.7 75 6. New America Foundation - Contextual Analysis of Community Network Sustainability

6.1. Introduction

Contextual factors such as regulatory and socioeconomic conditions have impact on the sustainability of community wireless networks, yet they are understudied compared to technical characteristics. New America’s contribution to the CONFINE project has been to develop a broad-based framework for understanding socioeconomic factors that determine the success, sustainability, and social impact of community networks; and to provide policymakers, as well as community networking advocates, researchers, and practitioners with a set of tools and best practices for building sustainability and positive impact into community networking practices. While our primary objective has been to understand the factors that shape sustainability for community networks, our secondary objective has been to support increased sustainability and social impact, including by increasing visibility of existing community networks and networking practices; lowering barriers to entry to broaden the base of participation and impact; and providing policy recommendations to support the social goals of community networking.

6.2. Background

As networked technologies transform the places where we live, digital participation becomes ever more important for access to basic rights, services, protections, and opportunities. While the academic and institutional ﬁelds of urban development and public policy have traditionally not included planning for communications systems, it is increasingly clear that broadband is an essential service, and that urbanists, governments, organizers, and residents all have roles in ensuring that everyone has reliable access. On the other hand, inequitably distributed broadband resources can have a profound effect on citizens’ ability to participate in economic and social activity.1 While most digital inclusion policies and programs address individual choices around broadband adoption and digital participation, the design and distribution of broadband infrastructure itself is a fundamental element of equity—and in our digital world, a question of equal opportunity and basic rights. Thus many policymakers operating in different sectors and roles are currently thinking about how to increase digital access and economic opportunity. What roles should government, private industry, and civil society play in ensuring that all citizens have access to sufﬁcient broadband services, and what can community-initiated networks offer? Can they provide a model of how to build more cooperative and resilient infrastructure with the engagement and leadership of local communities?

1According the the US Department of Commerce, up to 28% of American households, primarily in underserved and traditionally marginalized communities, currently do not have reliable broadband connections at home: http://www.ntia.doc.gov/press-release/2014/ digital-nation-report-shows-rapid-adoption-mobile-internet-use

76 6. NAF 6.3. Experiment Description

Community networking advocates have projected benefits—from digital inclusion to economic development—that should emerge by virtue of creating free public Wi-Fi access [43]. Yet in practice, some wireless projects have not lived up the rhetoric that has been employed to “sell” them to local governments and stakeholders. In the mid-2000s, many local community and municipal-community networks in the US failed due to issues around sustainability (in particular, lack of sufficient funds for maintenance and upgrades) and adoption (lack of interest or buy-in from intended beneficiaries) [44]. In these cases—for example in Tempe, Portland, Philadelphia, San Francisco, Santa Clara, East Palo Alto and other US localities where local actors tried to provide free public Wi-Fi in the mid- 2000s—failures of sustainability and adoption ended the projects, in some cases with millions of (public) dollars and goodwill lost along the way. Meanwhile, some community-led networks, like the constituents of CONFINE Consortium, are demonstrating a successful decentralized, cooperative “common-pool resource” approach to designing and building networked communications technologies. This phenomenon is not simply a fusion of private (corporate) and public (government) forces, but rather relies on community leadership, skills, and expertise. New America’s work with the CONFINE Consortium was designed to draw lessons from these existing, long-term and large-scale community networking efforts to examine their systems for alternatives to commercial models that have shown evidence of market failure—especially in the US, where corporate providers have failed to offer service that is sufficient or affordable for substantial sectors of the population. Further, we set out to show what kinds of social impact emerges aside from, or because of, Internet access via community networks. Finally, we wanted to understand who the “communities” are of these community networks, and whether networks with broader participation (beyond an “inner core” of tech hobbyists) show greater sustainability or social impact.

6.3. Experiment Description

As described, New America’s primary objective in our CONFINE research has been to identify the social and economic factors that contribute to the sustainability and social impact of community networks. As a confederation of robust community networks that have emerged independently in geographically and socially varied environments, the CONFINE project should provide an ideal opportunity for comparative study. However, New America researchers found uneven and ultimately insufficient interest in collaboration on the part of other CONFINE partners, and thus adjusted our methods to better serve all of our objectives (understanding and supporting sustainability for community networks by increasing awareness of existing, successful models; broadening participation; improving policy frameworks to leverage social impact; and lowering barriers to development of community networks) given the limitations that emerged. We will describe these challenges in greater detail in Section 8.5 below. Our original research plan included a multi-phased approach starting with geospatial analysis of CON- FINE networks in socioeconomic context; then proceeding through integration of the M-Lab diagnostic network measurement tool in order to understand the performance and uptake of CONFINE partners’ community networking services; then deepening findings with qualitative data-gathering in partnership with CONFINE members to build case studies around the development and evolution of successful community networks. In practice, our study used methodologies and produced deliverables in line with our original objectives; yet with extensive detailed empirical data drawn from only one of the CONFINE networks (Guifi), and with additional data drawn from US case studies. Following are details on our experi-

Deliverable D.4.7 77 6.4. Test Setup and Results 6. NAF ments using geospatial analysis techniques; regulatory, policy, and market analysis; scholarly analysis; and toolkits and best practices for application across the ﬁeld.

6.4. Test Setup and Results

6.4.1. Test Method 1: Geospatial analysis

Community networks do not take shape in isolation. They grow in the larger context of neighborhoods, districts, cities, and regions. It is important to understand how users, technologists, and organizers cooperate to build networks; understanding the physical and social environments where they work can also teach us about what kind of dynamics and practices best support sustainable projects. Diagnostic mapping can help illuminate how a network has evolved, by looking at expansion in different places at different points in time in relation to other factors. The purpose of this kind of analysis is to understand what conditions contribute to growth and sustainability. If we look at growth of the networks over time by using node databases showing dates of installation, we should be able to pin events in the timeline of a network’s growth to other patterns. In addition to helping us understand how networks have evolved, geospatial analysis can also help networkers plan deliberately for future expansion or for outreach and training. Internet connectivity needs and assets vary from place to place. Successful community networking projects account for those variations in their planning, either by tailoring a solution to a specific community or considering the placement of nodes as part of a larger regional plan or set of projects. Diagnostic mapping can help guide that process. For example, organizers or researchers may find that there are few nodes in an area with high population density; or, if they are interested in addressing the “digital divide,” they may wish to learn the locations of populations likely to experience challenges getting online, and focus efforts in those locations. Organizers working to create local community networks or training programs can also use mapping tools for network advocacy. Community leaders can use this approach to highlight inequity in digital access across neighborhoods, towns, or regions, and indicate the potential of community networking to address those gaps. We chose to begin our analysis by building a geospatial database for the Guifi network as an example of the process of community network mapping using data and software tools. We committed to using open software and data sets in line with openness and transparency principles. Using the open- source geospatial information system software QGIS, we layered Guifi’s node database over base maps from OpenStreetMap showing municipal boundaries and streets. We also added land use data from OpenStreetMap and demographic data from the Spanish census. Each layer contains data-rich geographic information which allows us to visualize and compare multiple points of information. As an example, we have created a set of choropleth maps (thematic displays in which areas are shaded or patterned in proportion to the measurement of the statistical variable) that contextualize the expansion of Guifi’s nodes with relation to population density. Using the mapping database, we can continue to work with these layers, add additional layers, or call out different variables to display on the map, depending on which demographic or other factors we wish to query. Based on our experience with mapping broadband access in the US, OTI initially proposed to map multiple networks to examine what contextual factors—such as demographics, institutional location/support, and other infrastructure—seemed to contribute to their growth over time. Based on these findings, we could posit and test ideas about the effect of different factors community network

Deliverable D.4.7 78 6. NAF 6.4. Test Setup and Results expansion. However, certain limitations regarding the availability of geospatial data presented significant challenges: 1. Incompleteness or inaccuracy of open and crowdsourced data sets. Data repositories like OpenStreetMap curate data volunteered from users. This means that data can be piecemeal—for example, open land use data is only available for about half of Barcelona. For complete data, researchers often have to purchase datasets from proprietary repositories or request them from government agencies. Crowdsourced contributions tend to be non-standardized, and cleaning the data for use was more complex than anticipated, especially when working in an unfamiliar language. 2. Variations in data format and coding practices. Community networking practitioners have particular requirements for data formats, and these are often non-standard. Guifi maintains a rich database of nodes, with information about installation dates, hardware in use, status of the node, what type of node it is, and so forth. This database is quite large, and is formatted in a custom XML format called Community Network Markup Language or CNML. The format is intended to be extensible, with a goal being that other networks would adopt a similar data specification. Unfortunately, it is not an easy format to work with in mapping software such as QGIS. This means that the data must be parsed and re-formatted as a CSV or other basic file to be readable by mapping software such as QGIS. This could also be remedied by Guifi and other networks providing data sets in multiple standard formats. 3. Unfamiliar coding of census data. Each national census bureau uses different codes for different kinds of geography. In the US, it is relatively common practice to pair the census data with geography files to create maps of the data. To join the data sets, you need to create an identifier that can exists in both data sets. In the US this is done by combining the state, municipality, county and census tract number together to form an 11- digit identifier. In Spain, it appears that working with geography and census data is not as common, and there are multiple sources for geographic data, such as a map file of census tracts. A similar ID is necessary to join the data sets, but there are also more combinations for creating the identifier due to the different ways that the Spanish states and municipalities group and re-group in different administrative contexts. This provided an extra challenge for understanding the census data in our test case. The geospatial analysis has provided us with some understanding of the demographic and social context of the Guifi network in Catalonia, and the same techniques can be applied to mapping other networks. This type of analysis could be very useful for sharing with policymakers or regulators who wish to understand the potential social or economic impact of community networking. However, Guifi has the most extensive available databases for geospatial work among the CONFINE networks, and even so we found that we did not have the resources to replicate this process with other networks. Rather, we have opened our methodology and approach for use by other researchers.

Network expansion over time

Guifi’s node database show the first nodes appearing in 1999, with a steep increase in number of nodes added per year (until 2012-2014). The following maps show all Guifi nodes installed from 1999 until August 2014. In figure 6.1 we show nodes that had “working” status in the database when we last downloaded the data. However, we do not know why or how often a node status may change, or how often. Further analysis could show us more about the location of “planned” nodes in addition

Deliverable D.4.7 79 6.4. Test Setup and Results 6. NAF to “working” ones.

Figure 6.1: All GuiﬁNet Working Nodes by Year Created

The most recent growth is on the coastline and in towns within commuting distance of Barcelona, with the most concentration in the Gurb region where the network was started. The following maps show snapshots of the development of the network over time. This first map (1999-2005) in figure 6.2 displays the initial establishment of the network in the Gurb area, consistent with Guifi’s founding narrative that Ramon Roca, a tech sector worker, started Guifi so that he would not have to go to Barcelona to get online, since commercial service was not available in the area. Relationships with local officials and community members allowed Roca’s group to gain access to rooftops and steeples for node installations.2 The following years showed network expansion southwest toward parts of Barcelona and some surrounding towns. In the first 7-year “incubation” period (1999-2005), only 372 total nodes were installed; in the following two years shown in the map in figure 6.3, 4783 nodes were added, a huge leap. The following years show an increased rate of expansion, with 5480 nodes added between 2009 and 2010 (shown in figure 6.4) and 7981 between 2011 and 2012 (figure 6.5). In the most recent map in figure 6.6, growth continues following the same pattern, and there are even a few nodes in Lleida and Girona. All other urban centers in Catalonia have experienced major growth. Overall, this case shows a pattern of relatively slow suburban/rural establishment followed by rapid growth in one particular urban area and its metropolitan region, yet very little growth in some urban

2 http://rising.globalvoicesonline.org/blog/2013/12/11/guifi-net-spains-wildly-successful-diy-wireless-network/

Deliverable D.4.7 80 6. NAF 6.4. Test Setup and Results

Figure 6.2: GuifiNet Working Nodes Created 1999-2005 centers. In order to further investigate this pattern, we can next start to examine differences between those urban centers that experience network expansion and those that do not. Factors that could be examined further in the Guifi case include: • Industry composition: does the Barcelona region have a greater proportion of industries that require Internet connectivity—for example, IT, media, and tourism—than Lleida and Girona, where much less activity occurred? • Demographics: Is Barcelona’s population younger or wealthier than that of Lleida and Girona? • Do regulations such as permitting and access to public buildings differ from city to city? • More targeted trigger points for growth: these maps break down growth by year, but we have much more detailed time series data from the Guifi node database and can look for the effects of specific events. Further, in geospatial analysis of other networks, it would be interesting to see whether an initial incubation period in a rural area, especially one without many commercial Internet service providers, improves the long-term sustainability of a network.

Network expansion in relation to population

The following thematic map in figure 6.7 shows the relationship between population density and the density of the Guifi network. It largely reflects the findings from the network expansion maps—that is, the network started northeast of Barcelona and has spread southwest to the Barcelona metropolitan

Deliverable D.4.7 81 6.4. Test Setup and Results 6. NAF

Figure 6.3: GuifiNet Working Nodes Created 2006-2008 area and along the coasts. The network has not expanded to some of the most populous areas of Catalonia, but rather in clusters mostly around the Barcelona and Gurb areas. Following, taking a closer look at the Barcelona area, it appears that network density does not always align with population density, as shown in figure 6.8. Further steps with this part of the analysis could include comparing demographics (age, income, educational attainment) and neighborhood composition in areas of the city that have been home to dense network growth. Do these neighborhoods host universities or other network building assets? Conversely, in areas with low network density, what factors have made those environments less conducive? Overall, this initial glance at the demographic patterns shows us that network density is not simply echoing population density, but following other patterns. This begs further research into the demographics and other factors that may help explain these patterns, but at this point local knowledge may be the most valuable source for information about the neighborhoods where the network has flourished and those where it has not. Without dedicated collaboration from a local partner, or greater access to data resources, we are unable to take the next steps in analyzing social patterns related to network growth. We hope that those with local relationships and resources are able to use this groundwork as a basis for further research.

6.4.2. Test Method 2: Regulatory, Policy, and Market Analysis

In order to develop an understanding of the policy and market environments shaping both the CON- FINE and other networking projects, we conducted a literature review on the social impact of documented community networking projects, as well as a series of interviews with CONFINE partners as

Deliverable D.4.7 82 6. NAF 6.4. Test Setup and Results

Figure 6.4: GuifiNet Working Nodes Created 2009-2010 well as other community networking practitioners and municipal officials.3 In examining empirical research studies chosen as explicit tests of the extrinsic and broader social benefits claimed for community networks, we found similar findings across different networks and scholarly disciplines. While social outcomes such as participatory community development and citizen empowerment do not emerge simply by virtue of the establishment of a community network, documented networks—including AWMN, Ile Sans Fil, and a previous generation of networks studied under the Fifth European Research Framework Programme (1998-2002)—have had success with: disciplining the broadband market; expanding access to underserved areas; fostering innovation communities; and demonstrating alternative service models and types of partnerships. Yet also across all case studies analyzed, the “community” of these networks tends to be insular—not inclusive of broader publics beyond providing Internet access to them—and, in fact, the inward-facing nature of these networking communities appears to be one key to their sustainability, as members are personally invested. In order for networking communities to keep from becoming insular and thus having limited socioeconomic impact for host communities beyond provision of access, we have concluded that network organizers must engage and strategically align their goals with broader publics and political bodies to collaborate on network design and sustainability planning. Alison Powell (2008) suggests that, in order to aim for social goals that are better aligned with the practices and needs of broader place-based

3Including: Bart Braem, Wireless Belgie; Ramon Roca, Guifi; Manos Dimogerontakis, Guifi and AWMN; Armin Me- dosch; Diana Nucera, Detroit Community Technology Project; Adam Longwill, Pittsburgh MetaMesh; Stu Jeffrey, WiFi 101; Debra Lam, Chief Innovation Officer, City of Pittsburgh; Joshua Breitbart, Special Advisor to the Mayor on Broadband, City of New York; Stu Jeffrey, WiFi 101 East Palo Alto

Deliverable D.4.7 83 6.4. Test Setup and Results 6. NAF

Figure 6.5: GuifiNet Working Nodes Created 2011-2012 communities of users, network organizers must create [45]: ...different kinds of collaborations to prevent new kinds of divides from forming between educated, professional users of WiFi and other people in the local community. . . As complex as the internal relationships may become, policy-makers and community organizers should attempt to leave space for visionaries, idealists, artists and geeks to think, talk, and hack their way into new publics. We hypothesize that there is an opportunity to change the insular nature of many networking communities by expanding the focus of “geek publics” beyond technology to include other affinity-based organizations that incorporate technology as a tool: for example, arts and media-making, community organizing and development, or resilience and preparedness. Following Maria Bina’s (2007) findings, one key to sustainability for community networks may in fact be participant self-selection and the enjoyment of crafting a platform for hobbyist knowledge- sharing—even if that knowledge sharing is deliberately cross-disciplinary and extends beyond facilitating access to other socioeconomic, sociopolitical, or cultural goals [46]. An intentionally developed social framework can provide historical context and tools for information-based community development in order to help networked publics self-define, set goals for broad and diverse social benefits in the context of past successes and failures, and potentially, build hybrid networked communities that collaborate to innovate. With the documentation of the CONFINE Consortium, practitioners have an opportunity to advocate for policies and investments that will support broad community benefits. Organizers who understand social outcomes can build on the proven ability of community networks to shape markets, expand access to underserved areas, and most importantly to explore new kinds of partnerships and collabo-

Deliverable D.4.7 84 6. NAF 6.4. Test Setup and Results

Figure 6.6: GuifiNet Working Nodes Created 2013-2014 rations in order to forge new affinity-based innovation and learning communities. To leverage that potential, community networks must be seen as a tool or platform in service of a broader vision defined by an intentionally defined (and intentionally inclusive) community—rather than building network infrastructure as a goal unto itself. Tools such as the technologies used by the CONFINE Consor- tium—software and firmware, wires, spectrum, and devices—are simply tools: it is the intention and composition of the organizers and innovators that makes the difference. In order to provide tools for collaboration to intentionally craft networking communities and shape their goals, we have also developed a series of planning tools and case studies available in the report Building Broadband Commons: Tools for Planners and Communities [47]. These include a methodology for mapping areas of need, specifications and recommendations for wiring large housing developments (including public and subsidized housing); and a popular education approach to training local residents (Digital Stewards) with varying levels of technical knowledge to become creators and maintainers of their own local networks and media. OTI has further developed of a github repository called the Community Technology Field Guide4, which includes curriculum modules and tools for groups to self-organize around the development of local community networks. Our CONFINE work partially supported and shaped this platform, and one CONFINE partner (Citizenship Square Kilometer) used our curriculum to implement network development for their CONFINE project. Our synthesis of documentation and findings regarding the social impact of community networks is forthcoming in a special issue of the Journal of Community Informatics (JoCI) on research methods for community informatics. Findings and extracts from both our policy paper and scholarly article

4 http://communitytechnology.github.io/

Deliverable D.4.7 85 6.4. Test Setup and Results 6. NAF

Figure 6.7: GuiﬁNet Working Nodes With Population have also been further published or quoted in The Atlantic magazine; a presentation at SXSW Inter- active 2015; and the online Coursera MOOC Technicity.

6.4.3. Test Method 3: M-Lab Integration

Measurement-Lab is a diagnostic tool designed to collect data on network performance using servers placed at distributed geographic locations to monitor anonymized Internet traffic over different networks. We intended to use it to understand the relative performance of CONFINE community networks, and especially to understand how community network offerings compared to the local market, and how performance affects network uptake and expansion. To establish a baseline dataset of previously collected network measurements for an example community network, we started with Guifi. We first identified the IP address ranges for Guifi, and then queried M-Lab’s historical data to find Network Diagnostic Tool (NDT) data submitted by users in the region. IP address ranges for Guifi’s various network locations using http://www.cityfreq.com/. For each IP address block, we constructed queries to find M-Lab data submitted from within these networks. Our first set of queries identified the total number of NDT tests initiated from Guifi IP addresses between January 1, 2013 and August 31, 2014. A total of 1150 NDT tests were initiated from Guifi IP addresses during this time. Table 6.1 shows tests by location: We then ran two additional queries to determine the download and upload throughput of these tests along several other sample variables, shown in table 6.2. Note that these queries are for completed NDT tests only and thus are a subset of the total attempted tests above. One location, Artes,´ returned

Deliverable D.4.7 86 6. NAF 6.4. Test Setup and Results

Figure 6.8: GuiﬁNet Working Nodes With Population, Barcelona

# of attempted NDT tests Guiﬁ Net City/Region 1/1/2013 – 8/31/2014 Avinyo,´ Catalonia, Spain 12 Artes,´ Catalonia, Spain 0 Barcelona, Catalonia, Spain 2 Gurb, Catalonia, Spain 36 Manlleu, Catalonia, Spain 40 Masquefa, Catalonia, Spain 654 Olost, Catalonia, Spain 92 Santa Eulalia, Catalonia, Spain 28 Tarragona, Catalonia, Spain 120 Torello,´ Catalonia, Spain 166 Vic, Catalonia, Spain 0 1150

Table 6.1: NDT tests by location no results. The data show a low number of tests in some areas, and slightly higher tests numbers in others. In all, the volume of measurement tests was extremely low, relative to other inquiries and analysis being undertaken by the M-Lab team. Also, due to privacy policies we could not parse the data to ﬁner geographies to observe network performance at a scale smaller than whole cities. Thus we suggested two options to improve the utility of the data to other CONFINE researchers.

Deliverable D.4.7 87 6.4. Test Setup and Results 6. NAF

Guiﬁ Net City/Region Avg. Median N Avg. Upload Median N Download Download Throughput Upload Throughput Throughput (Mbps) Throughput (Mbps) Avinyo,´ Catalonia, Spain 23.15 11.19 3 31.18 32.27 6 Artes,´ Catalonia, Spain ------Barcelona, Catalonia, Spain 52.57 36.66 18 42.50 33.76 19 Gurb, Catalonia, Spain 55.55 66.73 7 30.41 16.13 12 Manlleu, Catalonia, Spain 29.24 10.81 9 43.38 38.90 10 Masquefa, Mantmaneu, 27.80 26.11 100 34.10 38.43 100 Vilanova Del Cam, Catalonia, Spain, Olost, Catalonia, Spain 27.62 19.58 19 24.72 20.03 23 Santa Eulalia, Aragon, Spain 32.64 36.59 3 47.68 44.50 7 Tarragona, Catalonia, Spain 38.26 21.79 19 18.42 14.15 30 Torello,´ Catalonia, Spain 45.22 39.45 40 27.57 18.77 42 Vic, Catalonia, Spain 54.54 49.43 94 28.22 23.63 108 40.81 35.62 312 30.43 26.39 357

Table 6.2: NDT test results by location

First, increasing the number of user contributed and regularly submitted tests originating from networks could improve measurement of these networks and providers. Increasing test submissions could have been accomplished by a communications campaign encouraging network users to run existing NDT tests; however, this would have required cooperation from local network organizers, i.e. by sending out requests over their listservs to run NDT tests. Based on this option, we sent out a request over the CONFINE list, and created a flyer to be distributed via other CONFINE partners at the Rome convening in October 2014. However, we did not receive sufficient interest from other partners to engage in building a campaign with them. As a second option to increase data utility, Measurement Lab also maintains application program- ming interfaces (APIs) and sample “test integrations,” which allow network operators to integrate a localized version of NDT into a website, administrative portal, or software application. M-Lab engineers have prototyped test versions that can run on a small network connected device, which is configured to run tests at specific intervals, increasing the volume of test data and allowing better aggregation of the data. Thus, in partnership with CONFINE’s testbed administrators, OTI set up an instance of M-Lab for continuous automated monitoring of the Community-Lab nodes. Confine partners worked with M-Lab engineers to install a script to run automated NDT tests on a slice of the Confine Community-Lab testbed. M-Lab’s Network Diagnostic Test (NDT) on the federated Community-Lab testbed measures upload and download throughputs as well as minimum and average round trip times. This process tracks the health of the nodes themselves, and provides data for comparison to performance of other nodes in the individual community networks as well as potentially to nearby commercial service offerings. Data is aggregated of data out of BigStore (Google) to form a regular report query. Confine partners then use this report query to pull data submitted from the Confine testbed nodes. While OTI is not actively monitoring and analyzing these results, M-Lab infrastructure is providing an important documentation function.

Deliverable D.4.7 88 6. NAF 6.5. Main Achievements and Challenges

6.5. Main Achievements and Challenges

In our proposal, the Open Technology Institute at New America included the following caveat: “For researchers to successfully perform proposed research activities, community partners must be engaged and interested in participation.” Our research team found uneven interest in participation on the part CONFINE partners; while some were interested in collaboration, others showed less interest in studying the social dynamics of community networking. Overall, the level of coordination and engagement was not sufficient to expand some of our tests beyond an initial exploration of Guifi’s infrastructure and history, based on existing and open datasets. We hope that our methods can provide guidance to others who wish to conduct similar analyses, but have more complete access to data and local relationships. Indeed, we found some evidence of “research fatigue”—volunteers and members asked to answer similar kinds of questions repeatedly—among consortium partners. In many cases, community network gatekeepers (often volunteers) are asked to answer multiple and uncoordinated queries from different quarters. We could not find an instance of compilation or synthesis of previous work researching the socioeconomic context of community networks, aside from Armin Medosch’s helpful but as yet incomplete analysis. Our extended literature review searching for empirical evidence of social impact from community networking was thus a response to the lack of a unified knowledge base on these issues. However, while we hope that work is useful to future field researchers, we had difficulty conducting our work as originally proposed. While some of the problems we have encountered have been related to geographic distance, our review of existing literature demonstrates that in fact introversion and a reluctance to interact with “outsiders” is a more pervasive issue in the community networking space. As Alison Powell (2008) puts it, community wireless projects “create new potential for local community engagement, but they also have a tendency to reinforce geek-publics more than community- publics, challenging the assumption that community networks using technology development as a vector for social action necessarily promote greater democracy.” Similarly, Maria Bina (2007) demonstrates that while some community networkers talk about broader social benefits, their primary motivation is enthusiasm for working with technology alongside like-minded individuals. Community network advocates tend to talk about a wide range of possible social benefits without focusing on any in particular—thus, the actual progress toward external goals is limited. For these reasons, as documented in our midterm report to UPC (September-October 2014), we proposed adjusting our scope to develop tools to enable the expansion of the community networking public, in the hope that we can keep the community networking discourse from turning inward so much that it is irrelevant to the broader population. The comments of the EC reviewers at the last Confine plenary in Rome indicate that these question of larger social relevance are important to policymakers. As a UPC representative related5: They encouraged us to focus on socially innovative applications such as virtual curren- cies, adapt materials for non geeks, help understand why women don’t participate so much, look for alternative sources of funding, deepen on the evolution of economic/legal impact, think about how to communicate about our project, look at uniqueness of community nets and try find match with other social initiatives, how to enlarge and reach larger and different domains, support society in innovative ways, be examples for other initiatives and new social models in collaboration.

5Personal correspondence, CONFINE listserv 10/31/15

Deliverable D.4.7 89 6.6. Conclusions 6. NAF

While New America’s achievements in the course of our work with CONFINE diverge from our original scope, we have effectively • Increased awareness of existing community networks and their impact. Through publication and public presentation of our research, we have brought increased international visibility to the work of the CONFINE consortium. In addition, the lessons learned and best practices that emerge from the experience of CONFINE partners is contributing to an expanding pool of knowledge and community of practitioners. • Enabled broader participation in participatory, bottom-up and community-led networking activities from broader communities. Alongside greater awareness of community networking comes greater possibility of participation and support from local communities. If these networks can be understood as drivers of local economic development and the consolidation of social support networks, they will garner broader support and expand their potential impact. • Improved policy frameworks to leverage socioeconomic benefits of increased connectivity. A greater understanding of regulatory environments and of the markets and impact that emerge from more robust competition, including self-initiated, community-owned networks, helps policymakers create and support frameworks for more equitable and inclusive connectivity. • Lowered barriers to entry and greater chances of success for future community networks. As best practices models and information about how to interact with regulatory, market, and social contextual factors becomes more available – and as awareness of community networks increases – more local communities will be impelled to create and customize networks. In addition, more available information about these factors will invite in a greater variety of people with a wider set of skills, expanding the pool of community network enthusiasts beyond technologists to include entrepreneurs, organizers, artists, etc. Informed by the findings of our CONFINE research, New America is currently moving into leading the implementation of six new community networks in New York City. Further, we have provided consultation for technology and innovation officers and community foundations in Boston, New York, San Francisco, Silicon Valley, and Pittsburgh as they consider community networking models as part of an approach to municipal broadband.

6.6. Conclusions

A community-led method of planning and provisioning broadband employs local partnerships, can reframe the role of government and other institutions, and treats citizens as collaborators and experts on their own needs. As cities and towns work towards planning more collaborative, redundant, flexible, and ecologically adaptive systems in general, broadband infrastructure can be a site of pioneering cooperation. However, there are significant gaps in the field of community networking in availability of information about social impact and potential. While many advocates take an “if you build it they will come” approach to providing connectivity via community networks, there is a lack of evidence of broad social impact for community networks. Further, there are indications that the most sustainable community networks may also run by insular and heterogenous groups, which lack actionable plans to foster social outcomes. We encourage the Consortium, its members, and its reviewers to plan more deliberately for the integration of social research into future collaborative work examining the potential of community

Deliverable D.4.7 90 6. NAF 6.6. Conclusions networking. Without a more well-defined process for integration and collaboration, however, it will be difficult to generate valid findings and to work efficiently. Without structured and collaborative in- tent, the practices surrounding deployment, research, and use of community networks cannot become more inclusive and socially innovative.

Deliverable D.4.7 91 7. ICARUS - Impact of Community networks as Alternative infrastructure in remote and Underserved areaS

Providing affordable communication services to rural areas of developing countries has been a topic of concern in the last decades. The potential of community wireless networks to alleviate the cost problem, and to offer other development beneﬁts to their users, is well documented. However, very few exist in rural areas of developing countries, so the challenges to implement them and its impact are little known. This chapter introduces the case of a Community Network in rural South Africa and presents the results of a study to analyse the impact of connecting it to the Internet. The connection and the provision of this new service has been possible after obtaining a license-exemption from the regulator to do so. This has allowed the community to experience new services such as the breakout calls. However, this service has not been used that much and the Community Network has not had a direct impact yet on the patterns of and expenditure in communications of the people in Mankosi. Access to the Internet has been proven to have a major impact in the community with local students who managed to complete their application to a Higher Education Institution being granted a bursary to continue their education. Direct access to the Internet is not the only positive impact of the Community Network. Results show how it is being instrumental to develop and reinforce other social structures within the community.

7.1. Introduction

Providing affordable communication services to rural areas of developing countries have been a topic of concern in the last several decades. The lack of infrastructure (road and electricity) coupled with a sparse and low-income population have prevented conventional market forces from being involved in the solution [48]. Even thought mobile phone operators have widely increased their coverage in these regions due to the overwhelming and unexpected uptake of mobile phones by poor people [49], the problem is far from being solved. Communications remain unaffordable to most [50], given the increased cost, when available, of both voice [51] and data communications [52,53] that rural dwellers incur when compared to their urban counterparts. The potential of community wireless networks to alleviate the cost problem, and to offer other development beneﬁts to their users, is well documented [54, 55]. However, the lack of technical capacity to maintain and operate telecommunication networks in these regions have been pointed out as one of the main reasons that prevent these initiatives to reach a broader uptake [56]. The proliferation of urban community networks, where scarcity of spectrum, scale, and heterogene- ity of devices pose tremendous challenges to their stability and to that of the services they aim to provide [7], has fuelled the creation of robust low-cost low-consumption low-complexity off-the-self wireless devices which make the deployment and maintenance of these alternative infrastructures in rural areas [50] much easier. Thus, although more research is needed to continue solving the aforementioned technical challenges, existing technology can be enough to solve the affordability problems in remote and under served regions, where some of those technical challenges do not apply. Leaving the technical complexities aside allows us to focus on efforts to understand the socio-

92 7. ICARUS 7.2. Background economic effects of deployment. Considering that most community networks are deployed in urban settings, the few studies using a sound methodology focus on them [57–60]. This leaves government agencies, civil society organizations and the public in general with almost no scientiﬁc data to justify engaging in the deployment of community networks in rural areas.

7.2. Background

With an aim to close this gap, a Community Network project based on Ethnographic Action Re- search [61] commenced in April 2012. It started with a needs assessment in Mankosi, a traditional rural community, in the Nyandeni Local Municiapality, (O.R. Tambo District, Eastern Cape Province, South Africa), and resulted in the deployment in June 2012 of a bottom up community network to address the high costs incurred for communications [62]. The Eastern Cape is one of the most disad- vantaged provinces in the country, and the O.R. Tambo district has the lowest Human Development Index (HDI) (0.45) and the highest poverty gap (231 million) in the Eastern Cape. The number of people living in poverty is also high in this district (64.6%); unemployment is at 65.5% and the literacy rate 42, 2% [63]. Mankosi comprises 564 households in twelve villages that are spread across 30 km2 of very hilly land. Families of up to five adults and seven children live in homesteads: clusters of thatched, mud-brick rondavels, an occasional tin-roofed 2-room dwelling, an animal corral and a garden for subsistence crops. Households survive on less than US $200 per month, from government grants and payments from family members who temporarily migrate for work. Access to services is very limited (no tar roads, and grid electricity in only 2.1% of the households), as it is the access to higher education from its dwellers as only 13% of the population has completed high school or a higher level of education 1. Like 36% of South Africa’s population, inhabitants are governed by a Tribal Authority, which in Mankosi consists of the Headman, 12 Subheadmen and messengers. The Headman and Subheadmen’s homesteads are sites for local administration [64]. The community network (CN) intends to provide cheaper voice and data services in Mankosi. Before the participation of The University of the Western Cape (UWC) in the CONFINE project, the CN was allowing calls to be made for free among 10 points of presence in the community. These 10 nodes are installed in private houses that where selected locally, provided that they complied with some technical constraints [62]. The local nature of the calls was devised to avoid incurring recurrent costs like an Internet connection that would have allowed the provision of other services but which could have jeopardized the financial sustainability of the initiative. According to a needs assessment carried out in April 2012 a faster and higher uptake of the local service was expected [65]. However, data collected in June 2013 showed very limited use. Among the causes to justify this were local politics, the Tribal Authority and the inability to call mobile outside the network. Given the technical feasibility of providing that service (calling mobile phones), a socio-economic plan was devised. Based on the experience from [64] the initial design of the wireless station’s power supply included two cigarette lighter sockets to allow users to generate revenue from charging phones to help make the community network sustainable. Sine June 2013, more than US $1,000 have been collected collectively, with a consistent monthly contribution from all station holders. This money has been used to finance the Internet connection and the voice over Internet Protocol (VoIP) provider required to call mobile phones from the stations installed. However, these calls have a minimum cost that need to be, at least, recovered to keep the project financial sustainable. In South Africa, in order

1Data obtained in the Baseline, see Section 7.3 below.

Deliverable D.4.7 93 7.3. Experiment Description 7. ICARUS to charge for electronic services a license is required. So, in parallel to the money collection, a non-for-profit telecommunications cooperative was set up in March 2014 [66]. A member from each family hosting a station is part of the board of members. Apart from providing cheaper communication services, the cooperative has the objective of reinvesting all the revenue from the service provision for development projects decided locally. Once the cooperative was set up, they applied to obtain a license-exemption by the Independet Communications Authority of South Africa (ICASA), that was granted in September 2014. Although it was expected to have the license-exemption before this project started, this took longer than expected as it was the first time a similar application was made to the regulator. Despite the delay introduced in this project, it has set a precedent for other South African community networks to follow the same process. While waiting for the resolution from ICASA, we installed a 3G gateway and co-designed an IVR-based billing system that allows the cooperative to match the usage of the service with the invoices received from the VoIP provider [66]. So, since September 2014 the community network was connected to the Internet and making calls to landline and mobile phones was a real sustainable possibility. This allowed the researchers to focus on the main objective of this project: ”Provide scientific knowledge about the socio-economic impact of connecting to the Internet a community network located in rural South Africa in the lives of the community members and in the sustainability model of the service provision”

7.3. Experiment Description

In order to fulfil the objective described above, this project has focused on studying four concrete aspects of the Community Network in Mankosi: 1. The impact of the communication network on the communication expenditure and patterns of both mobile and non-mobile users. To do so, a panel study over three years was envisioned at the beginning of the project. A stratified sample of households from each of the 12 villages in Mankosi was surveyed twice before the intervention (first between July 2012 and January 2013, and then December 2013 and January 2014). A total of 255 questionnaires containing baseline data about households and individuals were collected. Due to inconsistencies discovered during the data cleansing process of the first questionnaire, in-depth interviews were conducted with the surveyors. Although the randomization of the households was done properly, cultural factors were overlooked. This prevented the proper randomization of individuals 2 inside each household and so the individual data is not representative of the population in Mankosi. Thus, only information from the Household Roster of the first questionnaire is used in this report. Dur- ing December 2014 and January 2015 a total of 75 people were followed-up and re-interviewed to see how telephone communication usage had changed after the implementation of the break out calls. In the last two rounds, surveyors used Open Data Kit to facilitate the data collection, including recording voice when open-ended questions were asked. 2. The spill over effects of the community network and the Internet connectivity. Ethnography and participatory observation have been the main sources used to unveil the spill over effects during the period under study. In particular, the main researcher has been living in the community 7.5 months during the first year of UWC participation in CONFINE. During this period he has attended 8 cooperative meetings, and 7 village meetings. This supplements the 9 months

2Cultural norms in the community includes addressing questions to the head of the household, in this case the eldest woman.

Deliverable D.4.7 94 7. ICARUS 7.3. Experiment Description

the main researcher lived in the community prior to that moment, when he attended other 15 meetings, including 7 village meetings and a community meeting. Additionally, while conducting interviews to understand the sense of ownership developed by ten of the main stakeholders involved in the in the community network, its perceived usefulness was assessed which showed additional spill over effects [67]. In this report, the study of the spill over effects are mainly focused on the board of members of the cooperative. A more comprehensive view on the whole community is being conducted as present. 3. The influence of the community network in the agency and aspirations of users: For the purpose of this project, it was decided to determine how the community network can assist youth from Mankosi to attain access to higher education. To do so, an Action Research intervention has been implemented as follows. The first phase started in May 2014 when a community forum was used to introduce this project and to gauge the amount of those who might be interested in being part of the study. A specific group was targeted: respondents between the ages of 15-35 who have at least passed grade 9. A database with potential respondents was created. From August 2014, the second phase of the fieldwork took place. First, workshops with six local researchers (ages 21-24) were held in an attempt to familiarize them with the aims and objectives of the project. The rationale was to provide them with the necessary training so that they were able to help in facilitating the workshops for prospective students during this project and in the future. All the local researchers had grade 12 as their highest qualification. Secondly, with the help of the local researchers, computer training with 30 prospective students were held between this time and December 2014. These trainings took place at the facilities of the local NGO, where a hotspot from the Community Network is providing Internet, so all the students where in one location to, firstly, optimize resources, and, secondly, to create a learning environment where they can be able to assist each other. XOs/OLPCs were used as the user device as they can be powered directly from a 12V source, similar to the one existing in the charging stations, to give the possibility to students to apply closer to their homesteads. Students were assisted to apply strictly to those institutions that are listed under National Students Financial Aid Scheme (NSFAS) because these are institutions that are funded by the government and therefore they provide bursaries to the financially needy students 3. Moreover, semi-formal interviews were held with provincial authorities as well as senior management of various higher education institutions within the province to determine existing initiatives and policies that are in place to facilitate this initiative. 4. The business model used to sustain the community network and its services. All expenses incurred and revenues obtained during the operation of the Community Network during the period under study have been collected. When possible, the economic benefits to users when compared with the services offered by traditional operators have been highlighted. Special at- tention is given to the intersection between cooperatives and community networks, and to the use of revenue from the system to other projects external to the community network. Addition- ally, solutions for other communities to start similar projects without having the initial capital expenditure covered externally are explored. In a socio-technical-economic experiment like the carried out during the participation of UWC in CONFINE, the longer the period of study, the more value the results obtained in this research may have. In this regard, and provided that we are not using the current facilities of the Community-Lab, we will continue gathering results until the last month of the project and they will be presented during the evaluation meeting.

3http://nsfas-ﬁnancial-means-test.software.informer.com/

Deliverable D.4.7 95 7.4. Test Setup and Results 7. ICARUS

7.4. Test Setup and Results

7.4.1. Study the impact of the communication network on the communication expenditure and patterns of both mobile and non-mobile users

Results from the baseline showed a high percentage of people using mobile phone services weekly (86.6%) and a high proportion of disposable income (21.97%) dedicated to a very constrained set of services (1.75 SMS and 4.35 calls, - 17.95 minutes - a week). Factors like mobile phone charging and mark-up of products in rural areas account for 23.24% of the total expenditure. Regarding data, 22.2% of the people access Internet monthly, but with very constrained use (25-30 MB a month). The results from the study of the baseline have been presented in detail in [68]. As said, a percentage of the total expenditure in communications is used for keeping the mobile phone charged. Considering the data reported in [51], people’s expenditure on charging has decreased over the past 3 years. In December 2010 they were expending R5.50 a week on average (R23.57 a month), while in December 2013 they reported to be expending R10.36 a month. This is a 56% reduction that can be explained by a combination of two factors: a) the installation by Zenzeleni Networks LTD of cheaper and closer charging stations (R3 vs R5.5 per charge); and b) more community members installing their own solar energy solutions to charge phones. Although there is no data to prove this as yet, qualitative findings suggest that the charging stations deployed as part of this project have also contributed to reduce the number of days without battery per month. Before the installation of the charging stations, a phone could be off for a week without being charged [64]. The importance for people to have their phones charged should not be disregarded, as shown by the number of days a month people report having their battery discharged (1.95 on average). Phones are not only used to make calls but they are used as a torch at night, for listening to the radio and music, and, provided they stay long periods of time without airtime, receiving urgent information from their relatives (news, job offers, etc). Then, it is no surprise that the feature people wish to have on their phones is a longer battery life (preferred by 58.0% of the individuals). A qualitative analysis on this is provided in the Section 7.4.2. Expenditure on airtime has increased by 33.05% when compared to the data collected by [51] three years prior to the data collected in the baseline. This corroborates the results obtained with regards to the correlation between years owning a phone and expenditure in mobile phone services (ρ=0. 215 (n=204), ρ¡0.01), a trend which is expected to continue in the future due to the even higher uptake of mobile phones in the past years. Provided there is a negative correlation between the number of days without battery a month and the airtime expenditure in a week (ρ=-0. 199 (n=199), ρ¡0.01), the installation of the charging stations (with its role in reducing the number of days without battery) may have contributed to the increased airtime expenditure. The comparisons between the baseline and follow-up study results of the 75 people that voluntary accepted to participate are presented herein. Although not significant, mobile ownership increased by 5% from the baseline (79%) to the follow-up (84%) surveys (p=0.2850; McNemar=1.14). Similarly, more people have bought their mobile phones from the baseline survey (76%) to the follow-up (83%) as opposed to use one given by someone else (p=0.2482; McNemar=1.3333). This trend is confirmed by the increased number of people who want to buy a new phone from the baseline (36%) to the follow-up (47%) surveys (p=0.1701; McNemar=1.8824). In both surveys longer battery life was singled out as the most important feature a mobile phone should have. It is worth noting that the percentage of people preferring a phone able to navigate the Internet has increased from 14% to 21% in between the two surveys.

Deliverable D.4.7 96 7. ICARUS 7.4. Test Setup and Results

Although not signiﬁcant, ownership of SIM cards has increased from 81% to 85% (p=0.3657; McNe- mar=0.8182). MTN was mentioned as the main service provider (91%) in both surveys. Although not signiﬁcant, network problems were experienced by 43% at the time of the baseline study, whereas this was slightly less (35%) during the follow-up study. Meanwhile, the number of people using someone else’s phone has decreased considerably from 11% to 1% (p=0.0196; McNemar=5.4444).

There was no difference in the reported monthly or weekly amounts paid for charging the mobile phone in the two surveys (p=0.9029; Wilcoxon Signed Ranks test (WSR)=5.5) . The number of days (in the last month) without a charged battery was similar in both surveys (p=0.8252; WSR=-18.5). No difference was found in the number of times per month people charged their phones in the two surveys (p=0.3587; WSR=-51.5).

Weekly airtime top-up patterns did not change significantly from the baseline to the follow-up study (p=0.2827; WSR=-68.5). Although not significant, slightly more money (mean=R3.32) was paid for the same airtime in the baseline study compared to the follow-up (p=0.2743; WSR=-92). There was also no change in the weekly upload of airtime on other people’s SIM cards in the two studies (p=0.8799; WSR=-9). The number of whole days without airtime was not different in the two surveys (p=0.7591; WSR=33). Social calling prevails as the main reason why people buy airtime (68.75%) in both studies. An increase on using airtime also for social Internet (from 4.17% to 10.42%) and a decrease on using it for SMS (from to 12.50% to 2.08%) from the baseline to the follow-up survey has been observed. Despite the similar expenditure patterns, there has been a significant decrease on the number of people acknowledging sacrificing on other things to buy airtime from the baseline (59%) to the follow-up (26%) (p=0.0002; McNemar=13.5). Still, from those answering positively, the item people forgo the most to buy airtime is the same in both questionnaires: essential food (44.4%).

The total duration of calls made in the week prior to the baseline survey was on average 6.6 minutes longer than that reported during the follow-up study (p=0.0428; WSR=-88). Although not signiﬁ- cant, the total number of calls in the week prior to the survey was slightly more (mean=-0.07) during the baseline study when compared to the follow-up study (p=0.8604; T-test=-7.5); and the average duration of each call was also slightly longer during the baseline study (mean=0.46) (p=0.3842; T- test=-39). During the baseline study, 25% indicated that none of the phone calls they made were to people living in Mankosi, whereas this increased to 42% in the follow-up study. Although not significant; there were slightly more “Please Call Me” messages send (mean=0.90) (p=0.5404; WSR=-68) and slightly more SMS messages sent (mean=0.45) (p=0.3309; WSR=-65) in the week prior to the baseline study when compared to the follow-up study. Together this may explain the higher airtime reported to be spent in the baseline study. The use of Internet applications on their mobile phones remained constant, at 28%, in both studies.

Although not signiﬁcant, during the follow-up study more people (96%) were aware of the solar charging stations compared to the baseline study (89%). In both studies, 62% indicated that they had used the solar charging stations. At the time of the baseline study, 87% agreed that the presence of the charging stations changed mobile phone usage patterns, whereas only 27% in the follow-up study still held this opinion (p=0.0004; McNemar=18). Perhaps many were at the time of the follow-up study more accustomed to using the solar charging stations and did not see it as a change in usage patterns anymore.

Although not a signiﬁcant change and given that only 21 people responded to this question, public phone usage increased signiﬁcantly from the baseline- (19%) to the follow-up study (24%)(p=0.5637, McNemar=0.33).

Deliverable D.4.7 97 7.4. Test Setup and Results 7. ICARUS

7.4.2. Collection and identiﬁcation of other non-expected effects of the community network and the Internet connectivity

The possibility of using the spare electricity from the solar system that is not used to power the telecommunications infrastructure has been one of the main non-expected effects of this community network. Apart from the impact on the communications patterns derived from a closer and more reliable source for charging phones [68] and its contribution to the financial model highlighted in Section 7.4.4, it has contributed notably as the incentive for the continued participation of the board of members of the cooperative. It was agreed with the community that in return for operating the charging stations and the community network for the community’s benefit, board members and their families could charge their mobile phones for free and light their homesteads using LED lights (provided they paid for the installation). Seven of the 8 houses where stations were installed belong to the 97.9% of the houses in Mankosi which do not have electricity [68], so they used this opportunity to light their houses and save on candles. The saving on charging and candles might be considered irrelevant on other cases but was capital to them and it was highlighted when they were asked about how they benefited from the initiative. As the chairperson of the board of members put it: ”Now, I am saving money, from charging phones for free and not buying candles and I can now have some money to buy things I need, like cabbage to make the food nicer. Now I am in no pressure to get electricity, because I have the lights. That’s make me feel proud.” Although there are other personal reason that may have influenced their availability to attend the cooperative meetings, the members from the two houses that do not host a charging station have attended to a considerable lower number of meetings. The role of electricity is not the only factor of the intervention affecting the degree of involvement of the board members. With increasing experience in decision-making and in internalizing the potential usefulness of the network, some of them have acknowledged that integrating the youth would be very helpful for taking the project forward due to their better understanding of technology. In a society where most decision are made by the elders the mere consideration of involving the youth is a considerable change of mindset. Despite an increase in the attendance to the cooperative meetings the board members’ children has been observed, their opinions, seldom voiced, have been barely considered. Their participation is in many cases related to their parents not being able to attend, but a considerable increase was experimented after the introduction of the breakout calls, when employment opportunities for those assisting with the process were discussed. However, as the use of the service never reached the levels expected, the youth interest, and so their involvement, decreased. The aspirations of the youth in relation of employment opportunities cannot be underestimated. Most of the local researchers hired to assist collecting data for this research project currently live in urban areas, either as a result of the Action Research intervention described in Section 7.4.3, or because employment opportunities are higher. Other unexpected effects observed go beyond the factors influencing the degree of involvement. The main local researcher that has assisted with this research is a major development actor in Mankosi. He has been working for the local NGO since 2008, and has been involved in other independent development and research initiatives taking place in the community. These were the main reasons behind his house being selected for hosting a charging station, and, so, him becoming part of the board of members of the cooperative. In addition, he is the only member of the co-operative speaking English, and having some previous knowledge about IT, so most discussions about the present and the future of the community network, and the transition in between, have been held with him. For all this reasons, other cooperative members have deposited on him the responsibility of arranging and

Deliverable D.4.7 98 7. ICARUS 7.4. Test Setup and Results facilitating the discussions and the decisions at the monthly cooperative meetings in an inclusive way. This is something new to the community as a board member explain: It is something that we are learning through this project. (...) In the project we are getting to know each other more deeply, like for example how I thought about X, and now my opinion has changed a lot positively. Something that we are achieving it that, people are working together. The CN is supposed to have gained better acceptance among the board of members because its values and approach resonated with local ways of being and thinking, such as respect, honesty and trust, and could seamlessly be emulated on local socio-political structures. It appears, therefore, that there was a localization of ownership through local structures, values and norms pre-existing the project. These structures and norms made easier the use and operation of collective infrastructures like a CN, as they already provided an unwritten basis for its “license”[69]. The lessons learned on this process has been applied to other development initiatives in the community. For instance, in the first stages of the project some of the committee members failed to attend the meetings. When asked for the reason they all mentioned that they had forgotten about them, and it was agreed that the lead LR would remind them two days and one day prior to the meetings. The LR had experienced similar situations in other projects (students attending the after classes courses, people failing to attend the meetings of other committees, etc) where the same solution has been applied successfully after the impact shown in this project. Similarly, the habit of taking notes, and reminding the decisions made in previous meetings have been applied to other committees where decisions have been consciously made to prevent inclusive participation and benefit a few selected ones even when the project is aimed to be community owned. The local NGO has also approached the researchers to replicate the process of nurturing and exercising community ownership developed around the community network [70]. Despite having been working in the community for more than 10 years, and trying to include locals in the decision-making process, the management remains in hands of workers and volunteers who are not from the community or share the same cultural background. In the last months, steps have been taken to use the experience of this initiative (described in detail [70]) to readdress the situation. The impact of these spill over effects is difficult to quantify quantitatively, and they might have had negative implications in the short term (like some committees having to restart their work from scratch). However, since the partnership between the community and the researchers was established the cooperative has met 25 times to coordinate its operation, which demonstrates the commitment to the project. A commitment that has not been achieved in any other initiative before. The process from passive to active entitlement of a historically disempowered community takes time [70], but it seems that this project might be contributed to it. As one member of the board explains: ”The most important thing that I have learnt is that I didn’t know that you could still get a little bit of money without having to go outside of the community to work hard. Just slowly, talking to people to set up the rules, getting a bit of money, starting a business that can help people in the future locally. So that’s a very big skill to me, as I didn’t know that we could start from a small thing and grow up.”

7.4.3. Analysis of the inﬂuence of the community network in the agency and aspirations of users

From having trained 30 prospective students, the number of those who applied for admission to a Higher Education Institution (HEI) drastically dropped to the 15 that did apply. This is as a result of the different costs involved in the application process, which are not reimbursed when the student receive its bursary, such: application fees which range from R100 to 400, registration fees which

Deliverable D.4.7 99 7.4. Test Setup and Results 7. ICARUS range from R500 to R5000 and travelling costs which range from R100 to R1000. Many institutions require students to send copies of their identity documents, copies of result and ID/Passport size photo via email; others require to print the application form from their websites, fill it in and then resend it back via emails or electronic forms, and others require to fax the required documentation. Due to the lack of equipment such as scanners and fax machines in Mankosi, travelling costs are necessary for students to complete their applications. Additionally, should they be accepted they need finances to travel from Mankosi to that particular institution. From the 15 students who applied for admission to HEI, 12 of them (ages 18 - 24) got admission and are currently enrolled. These students are registered for different courses across the spectrum, namely, Engineering – Civil (1), Mechanical (1) and and Electrical (3) - at Bufallo City College (BCC); Entrepreneurship and Business Management (1) and Computer Practices (1) at King Sabatha Dalindyebo (KSD); Project Management (1), Education (2) and Human Resource Management (1) at Walter Sisulu University; and Introductory Accounting (1) at Cape Peninsula University of Tech- nology. It is worth noting that, upon completion of their programs, students at BCC will receive a national Diploma, those registered at KSD a National Higher Certificate and those enrolled at a university a Bachelor’s degree. The reasons for declining the application of the other three students were: unable to pay application fees (2) and not having the sufficient average marks required (1). With regards to gender: from the initial 30 respondents who attended the workshops only 2 were males and from the 15 who applied for admission there was only 1 male. We have since put structures in place to figure out the reason behind these gender differences. All the students that got admission qualified for bursary and accommodation, even though it took longer for others to secure their bursaries as a result of outstanding documentation. For these students to qualify for bursary they have to meet two requirements. Firstly, they must have sound academic report or statement and secondly they undergo a Financial Means Test (FMT). This test determines the extent to which the student is financially needy. As expressed in Section 2, all these students depend mainly on a monthly social grant provided by the government to some members of their families. The amount of bursary given to each student depends on the cost of tuition fee, accommodation, food allowance, and traveling costs. The tuition fees differ from institution to institution and from course to course, as a result there is no stipulated or fixed amount of money that is allocated to each student. This bursary is not managed by the students; it is managed by the institutions in which they are enrolled. Students are given food vouchers quarterly and accommodation fees are paid directly to the landlord (in a case of private accommodation). No funds are transferred to student bank accounts. The students receive their bursaries in the first term of the academic year and all the students that were assisted have already received their bursaries. For the students to keep on receiving this bursary they must at least get an aggregate of 50 percent in their final results. Furthermore, students do not automatically qualify for this bursary and they need to reapply at the beginning of every year. When a student is applying for NSFAS, the following documentation is required: a proof of family income, birth certificate and identity documents of all the members of household, sworn affidavit and a proof of registration if there are any students in the household. These documents cannot be faxed or emailed, they can only be sent by post or be hand delivered. The process followed served to unveil additional challenges for the potential candidates to effectively enrol in tertiary education institutions: • Many of these students have never been exposed to a computer, so it was a huge challenge for them to grasp and understand the terms and instructions during training. For example, they did not know what a keyboard is, how to switch on the computer, etc. • Many of these students had to walk 5-9 km in a rainy weather to attend the training workshops

Deliverable D.4.7 100 7. ICARUS 7.4. Test Setup and Results

held at the local NGO Facility. The application process could have been done closer to the student’s houses using the hotspots of the Community Network, however, there was a dependency on equipment from the local NGO such as printer and photocopy machine. Still the applications had to be sent either via email (scanning documents), fax or post, but none of these devices were accessible from Mankosi. The interviews with provincial authorities and senior management of various higher education institutions served to unveil the challenges of HEI in the Eastern Cape Province: • 395 mud schools. • 3544 schools (14%) have no electricity supply. • 2402 schools (9%) have no water supply. • 19037 (77%) schools do not have a computer center, whilst • further 3267 have a room designed as a Computer Centre but are not stocked with computers. • 22938 schools (92%) do not have stocked libraries. • 21021 schools have no laboratory facilities. These challenges were reported to have an important effect on the learning process and on its output, preventing most students to meet minimum university entry requirements. Several interviewees agree that is not only the lack of resources, but the corruption, lack of capacity, incompetence, etc are the reasons for this situation. Some pointed at community organizations engaging with the state to report these cases and in order to make staff accountable as the solution to meet the right to education in the region.

7.4.4. Description of the business model used to sustain the community network and its services

The main premise for the business model used in the Mankosi Community Network has been to avoid offering a service without having the security that there was enough revenue to cover its running costs. The ﬁrst step toward this was an idea coming from the cooperative’s board members, based on experience from a previous project in the area that charged people’s mobile phones [64]. Provided there was spare power generated by the solar system, they decided to also use this spare capacity to charge the battery of mobile phones of fellow community members. The price decided was a 54.5% of the cost of using other local alternatives (R3.00 vs R5.50 [51]). Due to the lack of electricity in Mankosi (only 2.1% of houses have electricity [68]), this revenue stream has allowed the cooperative to collect R10,464 since the beginning of the project. A second idea was to offer breakout calls from the public phone. Each mesh potato in the network has a POTS handset attached; so, each is essentially a public phone. In order to provide a breakout service, an Internet connection was required that incurs recurrent costs. Prior to this, the calls were limited to within the network, for ’free’, and essentially avoids any running expenditure. With the money collected from charging mobile phones, a 3G gateway with an external antenna was purchased and deployed. Offers from satellite and terrestrial (offering microwave solutions) were studied but the cost of the infrastructure added to the recurrent monthly fees for the service were outside the budget of the cooperative. In addition to the low cost of the equipment, 3G could be used with data bundles, that, although increasing the cost per MB, allowed the cooperative to adapt incrementally with the demand and learn during the process. A billing platform was co-designed with the community in order to account for the revenue from the calls in a culturally sensitive way [71].

Deliverable D.4.7 101 7.4. Test Setup and Results 7. ICARUS

It is worth noting, that having the exemptions granted by ICASA has many positive sides effects apart from the obvious one of adhering to the South African regulatory framework and thereby avoiding prosecution. License exemption holders need not pay the registration fee, thus saving 2000 USD initial license fees (for ECS and ECNS) which is twice the money collected thus far in a year by the Zenzeleni for charging mobile phones. Additionally, Zenzeleni does not need to pay an annual percentage of the revenue obtained or the audit to calculate the fee. Avoiding these fees has direct economic beneﬁts, and also reduces the management burden as the holder does not have to produce additional documentation. The competition existing in the upstream VoIP market in SA together with the possibility to reach wholesale agreements with them has allowed the cooperative to purchase minutes at 1/6 (R0.20) of the prices offered by the Mobile Network Operators (R1.20). The cooperative has decided to resell them at 1/2 (R0.60) of those prices in order to have some margin to provide additional services. Despite this reduction, the public phones, now with breakout, have not had the uptake expected (only R112 collected this way after 6 months of operation) and the cooperative has asked the external researchers for help to study the feasibility to provide the service directly to people’s mobile phones, as opposed to the public phones only. Additionally, by obtaining the license exemption, different relationships are being explored with potential backhaul service providers, which could allow for the extra capacity required to provide VoIP service to mobile phones connected to the network, and to offer additional revenue streams such as In- ternet provision to other anchor tenants in the community (backpackers, school, and holiday houses) and, obviously, to the rest of the community. This process is resulting cumbersome, as incumbent providers are not used to work with license-exempt providers and an approval from their legal department needs to be granted. Zenzeleni is currently on the process to obtain that approval from the main provider in the country. It is worth noting, that the money collected is similar to the money expend (R10,842 vs R10,765). A breakdown of the expenses is presented in Figure 7.1.

Figure 7.1: Distribution of the co-operative expenses.

From this breakdown, two ﬁgures standout: • The amount of money used for transport and food (33% in total). This has been used to go to Mthatha, the closest city, to, mainly, do paper work (open bank account, obtain tax certiﬁcate,

Deliverable D.4.7 102 7. ICARUS 7.5. Conclusions

etc). This includes a percentage of the transport money (some of it is used for maintenance purposes). Once the transport used for maintenance is deducted, more than 26% of the money collected has been used for the institutionalization of the initiative. This amount would have been higher if the licence-exemption had not been granted by ICASA, which save additional R20,400 [66]. • The amount of money used (not considering equipment) for making the breakout calls possible provided their little use. Adding up the monthly fees paid to the VoIP provider to allow the service and the data bundles required for the Internet connection (which expire after 30 days) R670 have been spent. Last, it is worth noting that the initial capital expenditure (CAPEX) of all the infrastructure deployed in Mankosi (around R100,000), including telecommunications, solar and storage systems, was provided by the University of the Western Cape as part of the partnership. It is believed that if a business model able to cover the running cost of the CN, like the one described in this section is provided, it would be easier to access subsidies or credit to cover the CAPEX. One of this mechanisms could be the Cooperative Incentive Scheme (CIS) offered by the Department of Trade and Industry (DTI) [72].

7.5. Conclusions

The Mankosi Community Network is currently connected to the Internet after having successfully obtained a license exemption to provide electronic services. This has been possible by using the money collected from charging mobile phones to cover all the expenses incurred in the process (from equipment, to transport, to service providers). This has allowed the community to experience new services such as the breakout calls. However, this service has not been used that much and the Community Network has not had a direct impact yet on the patterns of and expenditure in communications of the people in Mankosi. Plans are in place to provide new services (Internet to anchor tenants and Inter- net to VoIP and Internet to WiFi enabled handsets), but challenges are being found with incumbent providers recognizing Zenzeleni as a service provider, and offering it wholesale pricing. These plans could allow contributing to reduce the outrageous percentage of their disposable income that people in Mankosi spend in communicating very little. Access to the Internet has been proven to have a major impact in the community with all the students who managed to complete their application to a Higher Education Institution being granted a bursary to continue their education. Most of the other student could not apply due to the lack of ﬁnancial means to face the application process. The cooperative is studying the idea of, once they start making some proﬁt, using it to subsidize student applications. Direct access to the Internet is not the only positive impact of the Community Network. Many results show how it is being instrumental to develop and reinforce other social structures within the community whose impact can be considerably more far reaching that the access to the Internet itself. More work is required to unpack these impacts as the project evolves.

Deliverable D.4.7 103 Part III.

Testbed expansion

104 8. CONFLATE - CONFINE extension towards OpenFlow experimentation: infrastructure, software and demonstrations

8.1. Introduction

The CONFLATE (CONFINE extension towards OpenFlow experimentation: infrastructure, software and demonstrations) project’s aim is the extension of Community-Lab in two dimensions: infrastructural and functional. The infrastructural extension is the deployment of new Community- Lab Research Devices, remotely available to researchers, within the Ninux.org network of Rome (and surroundings), which is the largest community network in Italy. The functional extension is the deployment of an OpenFlow eXperimental facility (OFX) within Community-Lab, which allows users/researchers to carry out Software Defined Networking (SDN) experiments based on the Open- Flow specification [10]. With the OFX extension, Community-Lab is, to our knowledge, the first wireless testbed which enables OpenFlow experiments within a large scale production network. OFX makes possible focused experiments on all OpenFlow key aspects: the switch, the controller, and the controller applications (i.e. switching rules). A researcher can deploy and test off-the-shelf or novel OpenFlow switches, controllers and controller applications. To simplify the realization of focused experiments, the researcher disposes of a pre-installed OFX-toolkit formed by: an OpenFlow switch (i.e. Open vSwitch) per virtual machine; an instance of an OpenFlow controller on a designated Community-Lab sliver; and a set of default controller applications, like L2/L3 Learning etc. The different items of the toolkit can be used or not, depending on the specific focus of the experiment. To demostrate the effectiveness of these extensions, we have performed an OpenFlow based experiment in Community-Lab using both the newly deployed Research Devices and the newly developed OFX toolkit. The experiment has a threefold aim: assessing the performance of different MPEG- DASH video streaming strategies in presence of Web proxies, serve as a guide for the usage of OFX, and pave the way to the deployment of an experimental video streaming service inside community networks.

8.2. Background

This section describes both how Community-Lab’s infrastructure has been extended through the deployment of new Research Devices in the Ninux.org community network, and how Community-Lab’s functionalities have been extended through the development and deployment of the OpenFlow Exper- imental Facility (OFX).

8.2.1. Community-Lab Testbed Expansion in Ninux.org

Ninux.org Rome is a wireless community network in Italy, born around 2002. Experimentation is one of its main goals: its members have been testing devices, auto-built antennas and routing pro-

105 8.2. Background 8. CONFLATE

Figure 8.1: Map of deployed Research Devices in the ninux Rome community network (left) and Research Device deployment example (right)

tocols while actively exchanging knowledge with the other Community Networks in Europe. Also, Ninux.org Rome is among the ﬁrst networks in Italy to provide1 native IPv6 connection to residential sites. It is an Autonomous System with IPv4 and IPv6 peerings with other local Internet Service providers and interconnected at NaMeX, Central Italy’s Internet Exchange Point. Many services are also provided by users inside the network: VoIP, XMPP chat, Web, search engines. Information about the network architecture and the employed protocols can be found in [73]. During the CONFLATE project, 20 indoor Research Devices have been deployed in the city of Rome and its surroundings at community network members premises, as depicted in Figure 8.1. Most of the deployed hardware are Intel Next Unit of Computing (NUC) devices with i3 CPU, 60GB of total disk capacity and 8GB of RAM. As in the Ninux.org community there are many technically skilled individuals, detailed instructions on how to autonomously prepare and deploy a Community-Lab Research Device and how to create slivers on them (both for research and community service provision purposes) have been published on the Ninux.org wiki [74][75].

8.2.2. OpenFlow Experimental Facility (OFX)

OpenFlow requires a layer 2 network and OFX provides a “L2 Virtual Topology Deployer” tool, which automatically deploys an arbitrary layer 2 topology among the virtual machines of the researcher. Obviously, this tool is also re-usable by other experimenters contributing to extend the Community-Lab experimental facility towards different kinds of layer 2 experiments, beyond Open- Flow. The layer 2 topology is formed by Ethernet tunnels. The tunneling approach makes possible a pervasive introduction of the OFX facility in Community-Lab, since it makes the OFX platform independent from the speciﬁc deployment of Research Devices (i.e., directly connected or connected through non-CONFINE devices) and from the underlying IP routing protocol used in the Community Networks (e.g. OLSR, BGP, etc.). Indeed, Community-Lab is rather heterogeneous with respect to these two aspects, therefore being the new facility independent from them avoids deployment issues. We planned to use the emerging network virtualization technology Virtual Extensive LAN

1with Unidata’s support

Deliverable D.4.7 106 8. CONFLATE 8.2. Background

Figure 8.2: VXLAN L2 Topology

(VXLAN) [76] for this task. At the time of the OFX toolkit development, VXLAN was not supported in the firmware running on deployed Research Devices, but newer firmware releases (based on OpenWrt Barrier Breaker) have added support. Therefore we provide two versions of the OFX topology builder: one based on VXLAN and one based on the well-known Generic Routing Encap- sulation (GRE) protocol, supported by all the CONFINE firmware versions. All the OFX source code has been openly published on the CONFINE Website [77] and are part of this deliverable.

8.2.2.1. VXLAN-based OFX topology builder

The Python script vxlan topology builder.py [77] automates the creation of VXLAN-based distributed L2 topologies by exploiting the following components: • the VXLAN Linux kernel module, to create VXLAN tunnels between stations and switches • the virtual switch module and tools OpenVSwitch, to create virtual switches supporting Open- Flow • the Paramiko SSH Python library [78], for remote node configuration The script connects via SSH to the nodes, creates tunnels between them and adds the virtual interfaces associated to the tunnels to local OpenFlow bridges. Figure 8.2 shows an example of a distributed VXLAN L2 topology along with the relevant nomen- clature that will be used in this section. The script vxlan topology builder.py has the following usage: vxlan_topology_builder.py [-h] [--clean] conf_file where conf file is a path to a JSON configuration file describing a given topology. If the clean switch is specified, the script will destroy all network devices and remove any configuration command.

Deliverable D.4.7 107 8.2. Background 8. CONFLATE

A “topology” JSON object consists of the following ﬁelds:

{ "topology" : { "id" : "example", "switches" : [ { "id":"sw1", "IpAddr":"192.168.0.167", "MgmtAddr":"fdf5:5351:1dfd:cafe:1001:0000:0000:babe", "SSHPubKey":"path_to_ssh_key", "SSHPassword":"root", "SSHUser":"root", "ports":[ {"ConnectedTo":"h1", "RealTunNIC":"pub0"}, {"ConnectedTo":"h2", "RealTunNIC":"pub0"} ], "OFControllerIP:Port":"192.168.0.254:6633" } ], "hosts" : [ { "id":"h1", "IpAddr":"192.168.0.168", "MgmtAddr":"fdf5:5351:1dfd:beef:1001:0000:0000:babe", "VIpAddrAndMask":"10.0.0.1/24", "SSHUser":"root", "SSHPassword":"root", "SSHPubKey":"path_to_ssh_key", "RealTunNIC":"pub0" }, { "id":"h2", "IpAddr":"192.168.0.169", "MgmtAddr":"fdf5:5351:1dfd:baba:1001:0000:0000:babe", "VIpAddrAndMask":"10.0.0.2/24", "SSHUser":"root", "SSHPassword":"root", "SSHPubKey":"path_to_ssh_key", "RealTunNIC":"pub0" } ] } } where: • id is a simple string that identifies the given topology • switches is an array containing a number of switch descriptions • hosts is an array containing a number of station descriptions The JSON object describing a switch consists of the following fields: • id: a string that identifies the virtual switch • IpAddr: the community public IPv4 address of the machine hosting the switch • MgmtAddr: the management address of the machine hosting the switch

Deliverable D.4.7 108 8. CONFLATE 8.2. Background

• SSHUser, SSHPassword and SSHPubKey: the SSH user password and public key for remote connection to the switch. Please note that if a public key is specified, a SSH password is not required (and viceversa) • OFControllerIP:Port: the IPaddress:port pair of an external Openflow controller • ports: is an array describing a set of stations and/or switches attached to the given switch ports. Each port object consists of the following fields: – ConnectedTo: the id of the station/switch connected – RealTunNIC: the real network interface to which the VXLAN tunnel is bound The JSON object describing a host consists of the following fields: • id: a string that identifies the host • IpAddr: the community public IPv4 address of the host • MgmtAddr: the management address of the host • SSHUser, SSHPassword and SSHPubKey: the SSH user password and public key for remote connection to the host. Please note that if a public key is specified, a SSH password is not required (and viceversa) • VIpAddrAndMask: is the IP address and netmask in the overlay network address space attached to a VXLAN device. If this is specified, all packets routed toward this network will automatically delivered to the VXLAN driver • RealTunNIC: the real network interface to which the VXLAN tunnel is bound The difference between a switch and a host is that a switch runs an openvswitch bridge which contains two or more tunnel endpoints, while a host has only one tunnel interface that is connected to a switch. To run an OpenFlow experiment on a distributed L2 topology built with vxlan topology builder.py an OpenFlow controller running on a (which can be the same machine running this script, or a sliver) might be needed. We have chosen the RYU Openflow controller [79]. In the OFX git repository [77] sample topologies (3 of which are depicted in Figure 8.3) are provided. The steps to run an Openflow experiment are: • write a RYU Openflow application. The RYU distribution has several sample applications in the directory ryu/app/. In this example we choose simple switch.py which is a simple Openflow application emulating 802.1 MAC learning operations. To start the controller run from the ryu root directory: # ./bin/ryu-manager --verbose ryu/app/simple_switch.py • run the CONFLATE topology builder with a given JSON configuration file: # python vxlan_topology_builder.py app-1sw-openflow.json • use ping to test the topology deployment. Please note that the machine running the Openflow controller and vxlan topology builder.py can be the same machine.

8.2.2.2. GRE-based OFX topology builder

The GRE (Generic Routing Encapsulation) based CONFLATE topology builder has been implemented in the gre topology builder.py script in the OFX repository [77]. It uses Layer 2 GRE tunnels (gretap in Linux terminology) to build an overlay topology on which OpenFlow tests can be run. Layer 2 GRE tunnels are supported by all versions of the CONFINE ﬁrmware.

Deliverable D.4.7 109 8.3. Experiment Description 8. CONFLATE

Figure 8.3

The GRE-based OFX topology builder uses the same JSON conﬁguration ﬁle format as the VXLAN- based OFX topology builder described above, and it can be used in the same way. Invocation example: python gre_topology_builder.py app-3sw-management.json

8.3. Experiment Description

To test the effectiveness of the proposed extensions, CONFLATE has used the newly deployed Re- search Devices of Ninux.org to perform a simple (but practical) OpenFlow based MPEG DASH Live Video Streaming experiment. Among the plethora of possible test applications, CONFLATE has selected video streaming since it is actually one of the killer applications in Community Networks and it is not (yet) available in Ninux.org. Moreover, the test setup serves also as a hands-on user guide about how to use the OFX facility to setup OpenFlow experiments. MPEG-DASH (Dynamic Adaptive Streaming over HTTP) is an adaptive video streaming technique. The video is encoded at different coding rates and divided into segments which contain a short interval of playback time.The segments can be made available as ﬁles hosted on a plain Web server. MPEG- DASH clients download segments using HTTP and during playback can select, based on current network conditions, the coding rate of the segments to download. Clients select the coding rate of video segments by using an Adaptive Bit-Rate (ABR) strategy; in particular, a throughput-based ABR is used in the case of ”live-video” applications. A client continously estimates the download rate and

Deliverable D.4.7 110 8. CONFLATE 8.4. Test Setup and Results on the base of this information promptly selects the suitable coding rate. MPEG-DASH has been ratiﬁed by ISO in April 2012 [80]. Recent work in the literature has raised the issue of video streaming performance in presence of Web caches [81][82] when employing client-side throughput-based ABR strategies. At the same time Web caches are of particular interest in case of wireless community networks, since they can reduce the end-to-end path from data provider to client: indeed, the cache is usualy closer to the client than the origin server. We use the OFX platform to compare the performance of different ABR strategies proposed in the litterature in an MPEG live-video streaming scenario in presence of a Web proxy within the Ninux.org network of Rome. We implemented four Python-based clients with the following ABR strategies to be compared: • Buffer-based (BB): a single step upscale occurs when the playout buffer reaches 3/4 of its capacity and a single step downscale occurs when the buffer goes below 1/3 of its capacity; • Throughput-based (TB): the coding rate is selected as the highest one lower than the average download rate; • DASHJS-based (DJ): we used the ABR rules of the dash.js v1.3.0 [83] reference player, namely: ThroughputRule (TR), BufferOccupancyRule (BR), InsufﬁcentBufferRule (IBR). These rules are concurrently used. The latter two have priority with respect to the former. TR is like the previous TB one. As a consequence of BR, the coding rate switches to the maximum one when the buffer is larger than a RICH-threshold (3/4 of playout buffer). Due to IBR, the coding rate switches to the lowest one when the playout buffer is empty; • PROBE-based (PB): it is the ABR logic of [82]. It is a TB strategy, but before performing an upscale to the TB coding rate, the client probes the server-client bandwidth to verify the actual availability of such a rate.

8.4. Test Setup and Results

This section describes both setup and results of the OpenFlow-based MPEG-DASH video experiment described in Section 8.3. To perform the experiment we employ the CONFLATE extensions to Community-Lab by selecting Research Devices deployed in Ninux.org and exploiting the OFX toolkit.

8.4.1. Test Setup

We set up the experiment by employing the OpenFlow Experimental Facility (OFX), using the GRE- based OFX topology builder tool (§ Section 8.2.2.2) to create an overlay OpenFlow network. The test scenario is depicted in Figure 8.4. We have allocated a slice in Community-Lab with 8 debian slivers: • 5 slivers act as Clients. We have uploaded on them the set of fake video players that implement the strategies to be compared. These slivers are not part of the overlay OpenFlow network, i.e. represent normal Ninux.org users • 1 sliver acts as a Web Server. It provides the MPEG-DASH video segments through a plain Apache HTTP server [84], and runs Apache Tomcat [85] with the servlets that are needed for the operation of the Probe-based (PB) strategy

Deliverable D.4.7 111 8.4. Test Setup and Results 8. CONFLATE

Figure 8.4: Experiment Scenario

• 1 sliver acts as a Web Proxy. It runs the squid daemon [86], in transparent mode, with caching and the collapsed-forwarding option activated. Both the requests directed to the plain Apache HTTP server and apache-tomcat servers described above are intercepted and processed • 1 OF Edge Switch sliver, which acts as a gateway to the OpenFlow overlay network, which includes also the Web Proxy and Web Server slivers. The overlay subnet (i.e. the IPv4 addresses assigned to OpenVSwitch bridges running on the slivers) is announced in the OLSR routing plane (as an HNA) by a community device connected to the OF Edge Switch sliver. The Layer 3 topology of the allocated slivers is schematized in Figure 8.5. To set-up the overlay OpenFlow network we have produced a description of it in the JSON format (communitylab-real-2.json in [77]). The devised network has a full-mesh topology to allow the simple handling of broadcast ARP requests. This description has been provided as input to the GRE-based OFX topology builder tool that has created a set of GRE tunnels between the slivers and initialized the OpenVSwitch daemon, adding the interfaces associated to the GRE tunnels to the OpenFlow bridge. Then we have installed some basic OpenFlow rules to provide basic IP connectivity between slivers. To this aim we have devised a Python script (basicopenflow.py in [77]) that takes as input the same JSON description provided to the GRE-based OFX topology builder tool. These rules have low priority and thus can be easily overridden by installing OpenFlow rules with higher priority. On the OF Edge Switch sliver we have added a flow modification rule, with higher priority than the basic rules, to redirect all IP packets that have as destination the server’s IP address through the GRE tunnel to the sliver hosting the Web Proxy. To achieve this task, we have used the OpenVSwitch tool ovs-ofctl, but the same flow rules could be easily pushed by an OpenFlow controller. On the Web Proxy sliver, which runs the squid daemon, we have enabled transparent mode by adding the DNAT iptables rules (which are normally employed for transparent mode Web caching) for all TCP segments having as destination ports 80 and 8880, which are the Server ports on which the apache server and the apache-tomcat are listening, respectively. Moreover, we have added a flow modification rule, with higher priority than the basic rules, to change the destination MAC address of the packets directed to the Server with the one of the OpenFlow LOCAL port, and then output them to the local OpenFlow bridge.

Deliverable D.4.7 112 8. CONFLATE 8.4. Test Setup and Results

1 0 .17 6 .0.2

client1 172.16.177.11

192.168.178.100

172.16.176.2

1 0 .16 2 .7.1

192.168.178.200

OF edge 172.16.159.26 1 0 .15 9 .1.66

10.159.1.251

10.162.0.225

10.162.0.221

client3 172.16.177.10

172.16.172.11

172.16.172.14

172.16.200.67

172.16.159.25

1 0 .17 2 .0.1

172.16.176.3

proxy 172.16.159.20 172.16.133.11

1 0 .13 5 .6.1

172.16.172.13

172.16.176.9 172.16.135.10 client5

172.16.139.251

10.133.10.254

1 0 .13 5 .2.5

172.16.139.252 client2 server

172.16.132.4

172.16.132.3

1 0 .13 2 .6.14

172.16.132.14

172.16.132.7

1 0 .13 2 .6.7

172.16.10.193

172.16.10.192

10.132.15.15

172.16.132.15

1 0 .13 2 .15 .1 client4

Figure 8.5: The topology of the network used for the experiment, inferred through the execution of the traceroute command on the slivers. Each node in the graph represents a Layer 3 hop. The round intermediate orange nodes represent Community Devices, while the rectangular nodes represent Community-Lab slivers. The ”OF edge”, ”proxy” and ”server” slivers are part of an OpenFlow overlay network, built using OFX.

On the Client slivers we have installed the Python scripts that implement the video players described in Section 8.3, and we have disabled reverse path ﬁltering (rp filter value in the proc ﬁlesystem is set to 2) to allow the download of video segments through asymmetric paths. We streamed in live mode the Big Buck Bunny movie [87], encoded with six representations, whose coding rates are 0.55, 1.5, 2.5, 3.5, 4.5, 8.6 Mbit/s. Segment duration is 4s, the number of segments is 160. The tests have been ran using Community-Lab nodes inside the Ninux.org community network over a timespan of approximately 42 hours, at different times of the day. The test results are illustrated below, in Section 8.4.2.

Deliverable D.4.7 113 8.5. Main Achievements and Challenges 8. CONFLATE

8.4.2. Results

Figure 8.6 illustrates the experiment results. The histograms on the right column represent the average of the 14 points shown on the corresponding plot on the left column. For a description of the compared strategies, please refer to Section 8.3. The strategies that have shown the lowest number of lost MPEG-DASH video segments are the TB and PB strategies, as depicted in Figure 8.6i and Figure 8.6j. Among these two strategies, the TB strategy achieves a higher bitrate (Figure 8.6c and Figure 8.6d), which corresponds to both a better quality of the streamed video and a higher server load (Figure 8.6e and Figure 8.6f), the latter miti- gated by a good Web cache hit ratio (Figure 8.6g and Figure 8.6h). The PB strategy has instead shown to be much more stable with respect to coding rate changes (Figure 8.6a and Figure 8.6b), and this is a rather important aspect to obtain a valuable QoE. Although the DJ and BB strategies have reached the highest bit rates (Figure 8.6c and Figure 8.6d), a loss of a signiﬁcant number of segments was observed (Figure 8.6i and Figure 8.6j), which corresponds to missing frames in the video stream. Compared to the DJ strategy, BB has shown a better cache hit ratio (Figure 8.6g and Figure 8.6h), which might be a consequence of the lower number of bit rate changes (Figure 8.6a and Figure 8.6b).

8.5. Main Achievements and Challenges

We have achieved an infrastructural and functional expansion of Community-Lab with the collaboration of the Ninux.org community. About 10 out of 20 Research Devices deployed inside the Ninux.org community network have been prepared, configured, and later maintained, by technically skilled community network members, following the guides that we have prepared [74][75]. This distributed and knowledge-sharing approach is in-line with the Ninux.org community culture and goals. A challenge was represented by community network members moving out of their houses during the project’s timespan, requiring a relocation of some Research Devices in other parts of the community network. The functional expansion of Community-Lab through OFX has allowed us to successfully deploy an OpenFlow-based experiment in Ninux.org. The OFX toolset has been openly published on the CONFINE Website [77]. In the development of OFX, a challenge was represented by the lack of support for Linux VXLAN modules in the CONFINE firmware, but we were able to circumvent the issue by successfully employing GRE. During the experiment preparation, we faced some network instabilities, which have to be taken into account when dealing with any real-World best-effort network. We also faced the need to have a network prefix announced inside the community network routing plane. The issue was resolved by asking a community network member to add the network prefix to his node’s OLSR configuration, but having a partial integration of Community-Lab with the routing plane of the community networks would be helpful in these scenarios. The devised experiment has allowed us to compare the performance of different state-of-the-art adaptive bit rate strategies for MPEG-DASH video streaming. Moreover, the proposed experiment can be expanded to build a community network service with a CDN-like architecture by adding more Web proxies and an SDN controller. The SDN controller would decide on which Web cache the video segment requests have to be directed, based on e.g. client-cache distance to minimize latency or maximize hit ratio. This would also mean that the Community-Lab platform could be employed beyond

Deliverable D.4.7 114 8. CONFLATE 8.5. Main Achievements and Challenges

60 60 DJ TB BB 50 PB 50

40 40

30 30

20 20 number of rate changes rate of number changes rate of number

10 10

0 0 0 5 10 15 20 25 30 35 40 45 DJ TB BB PB time (hours) strategy (a) number of rate changes, average of all 5 clients (b) number of rate changes, average of all test runs

9 9 DJ TB 8 BB 8 PB 7 7

6 6

5 5

4 4 bitrate (Mbps) bitrate (Mbps) bitrate 3 3

2 2

1 1

0 0 0 5 10 15 20 25 30 35 40 45 DJ TB BB PB time (hours) strategy (c) bitrate, average of all 5 clients (d) bitrate, average of all test runs

1000 1000 DJ TB BB PB 800 800

600 600

400 400 server load (Mib) load server (Mib) load server

200 200

0 0 0 5 10 15 20 25 30 35 40 45 DJ TB BB PB time (hours) strategy (e) server load, average of all 5 clients (f) server load, average of all test runs

100 100 DJ TB BB PB 80 80

60 60

40 40 hit ratio (%) ratio hit (%) ratio hit

20 20

0 0 0 5 10 15 20 25 30 35 40 45 DJ TB BB PB time (hours) strategy (g) cache hit ratio, average of all 5 clients (h) cache hit ratio, average of all test runs

120 120 DJ TB BB 100 PB 100

80 80

60 60

40 40 number of lost segments lost of number segments lost of number

20 20

0 0 0 5 10 15 20 25 30 35 40 45 DJ TB BB PB time (hours) strategy (i) lost segments, average of all 5 clients (j) lost segments, average of all test runs

Figure 8.6: Results of the MPEG-DASH client strategy comparison experiment ran in Ninux.org using Community-Lab with the OFX extension

Deliverable D.4.7 115 8.6. Conclusions 8. CONFLATE pure research purposes, to deploy experimental services in community networks.

8.6. Conclusions

We have expanded Community-Lab in two dimensions: infrastructural and functional. The infrastructural extension was achieved through the deployment of new Community-Lab Research Devices in the Ninux.org community network in Rome. The functional extension, instantiated in the OpenFlow eX- perimental facility (OFX), allows now researchers to deploy OpenFlow experiments in Community- Lab. To test the effectiveness of OFX, we have performed an OpenFlow-based experiment employing slivers inside the Ninux.org community network, in which we compare the performance of different Adaptive Bit Rate (ABR) strategies for the download of MPEG-DASH video segments. The results of the experiment show that none of the considered strategies has achieved the best performance with respect to all the proposed metrics. The described experiment scenario could serve as the basis for a CDN-like and location-based video streaming service to be deployed inside community networks.

Deliverable D.4.7 116 9. Sarantaporo.gr WiFi Networks

9.1. Introduction

The goal of Sarantaporo.gr Non Proﬁt Association’s participation in the CONFINE project under Open Call 2, was to provide expansion of the existing project’s testbed for enriching the experimental and testing environment. The Sarantapro.gr Community Wireless Networks, which was extended within the project, is located in a rural area of the Central Continental part of Greece – Sarantaporo village – and fourteen surrounding villages, located in the Elassona Municipality. Being very speciﬁc with its rural character and a socio-economic status of underdeveloped region, with no other alternative for modern connectivity in almost half of these villages, the Sarantaporo area was a suitable case to be added to the Community-Lab testbed in terms of diversifying the characteristics under research of community wireless networks.

9.2. Background

Sarantaporo is a small village in northern Greece located in the remote, mountainous area of mount Olympus. Similarly to many isolated areas of Greece it lacked Internet connectivity until 2010, when a group of local young people decided to address the problem and gap the digital divide that separated their village from the rest of the country’s IT infrastructure. Taking advantage of a GFOSS (Greek Free/Open Source Society) open call (Open WiFi) to build local community wireless networks, the newly formed group, named ‘Sarantaporo.gr’ after the village’s name, set to deploy a local Community Wireless Network (CWN) in the village. The equipment, over 160 OpenMesh devices (mesh routers based on openwrt firmware), was sponsored by GFOSS and the installation was made voluntarily by members of Sarantaporo.gr. Once the infrastructure was in place, a 3G dongle was initially used as an Internet gateway, due to lack of landline Internet connections. At the same time the villagers applied en masse for landline Internet connections, in order for a specific TelCo to be motivated to invest in relative infrastructure. This worked and a few months later ADSL connections were available in Sarantaporo village. A few of these were eventually used to connect the local CWN to the Internet. The OpenMesh hotspots provided the center of the village and nearby neighbourhoods with open Internet connectivity for locals and visitors alike. Soon, other villages of the area addressed the group asking for their help in setting up local CWNs in their villages. Within the next 3 years, until 2013, the project was replicated in a total of 15 villages. The final infrastructure comprised 160 access points. In villages where no Internet landline was available, a wireless point-to-point connection was created with Sarantaporo village, where a dedicated ADSL connection for each of these villages was added to the infrastructure. The whole effort is a bottom-up initiative, supported by local people, who volunteered hundreds of man-hours. The cost for the ADSL connections was covered by Sarantaporo.gr, local cultural associations and active citizens. Moving forward from the scope of providing WiFi Internet connectivity and on to the enhancement of social interaction and development in the villages of the area, the Sarantaporo.gr group completed,

117 9.3. Test Setup and Results 9. Sarantaporo.gr in the Summer of 2013, the compilation of a report about interconnecting the villages’ local CWNs via a backbone into a new bigger CWN. This effort was backed by important stakeholders, such as the Region’s Technical Institute, the Greek Free/Open Source Society (GFOSS), some local cultural associations, the Elassona Municipality and active citizens. Eventually this led us to found the Sarantaporo.gr Non Proﬁt Organization and apply for the CON- FINE Open Call 2, as a testbed expansion project. This was a unique win-win opportunity, for both the CONFINE project and the local communities. Being very speciﬁc with its rural character and a socio-economic status of underdeveloped region, with no other alternative for modern connectivity, the Sarantaporo area was a suitable case to be added to the Community-Lab testbed in terms of diversifying the characteristics under research of community networks. At the same time the project offered Sarantaporo.gr CWN a chance to inter-connect the villages WiFi “islets” via a backbone and also expand to the Region’s Technical Institute and through it to the rest of the European CWNs which already participated to the CONFINE project.

9.3. Test Setup and Results

In what follows we give an overview of our test setup and the results from the described concept.

9.3.1. Technical aspect

Sarantaporo.gr Non Proﬁt Association’s role to the CONFINE project was to expand the testbed with additional nodes and respective research devices. According to the initial planning and proposal, a minimum of 15 new backbone nodes would be added to the testbed. Within a time-span of 12 months, a total of 21 backbone nodes were eventually deployed, connected via 27 point-to-point wireless connections. The backbone interconnected 11 out of 15 local networks and federated the newly formed Sarantaporo.gr CWN (backbone network) with the CONFINE testbed. 18 research devices (RD) were installed in respective nodes. Four of the above backbone nodes where deployed in locations other than the villages:

• Larissa: at the Technical Institute of Thessaly • Giannouli: a suburb of Larissa • Elassona: the nearest town to Sarantaporo village (30km) • KEOAX: a military camp on the mountainside of Olympus mountain

In table 9.1 a complete list of the backbone nodes and research devices is presented. A rack and UPS power support unit were installed in every RD node as shown in ﬁgures 9.1 and 9.2, to provide for equipment protection and secure seamless operation. VMWare ESXi virtual machines were used to allow testbed integration to take place in parallel with production use. A wireless connection to the Thessaly Technical Institute provided the gateway to the Greek Research and Technology Network (GRNET S.A. www.grnet.gr), and via VPN the Sarantaporo.gr CWN was federated with AWMN and the rest of the testbed CWNs. We also reserved two ADSL lines for backup in case of failure of our main internet gateway.

Deliverable D.4.7 118 9. Sarantaporo.gr 9.3. Test Setup and Results

Table 9.1: Backbone nodes and research devices of Sarantaporo.gr CWN

Node Code Node Name Active connections Research dev. installed? #18 Melouna 6 Yes (Node ID:267) #1 Sarantaporo-Lia 5 Yes (Node ID:254) #66 Sarantaporo-Achilleas 5 Yes (Node ID:253) #67 Sarantaporo-Makris 5 Yes (Node ID:255) #3 Dolihi 3 Yes (Node ID:257) #8 Farmaki 3 Yes (Node ID:261) #14 Mikro Eleftherohori 3 Yes (Node ID:264) #2 Kokkinogi 2 Yes (Node ID:258) #5 Lofos 2 Yes (Node ID:259) #6 Tsapournia 2 Yes (Node ID:262) #7 Milea 2 Yes (Node ID:260) #9 Pythio 2 Yes (Node ID:256) #10 Flampouro 2 - #68 Sarantaporo-Ekklisia 2 Yes (Node ID:266) #69 Sarantaporo-Vasilopoulos 2 Yes (Node ID:270) #76 KEOAX (Military camp) 2 - #4 Kokkinopilos 1 - #19 Larisa (Technical Institute of Thessaly) 2 Yes (Node ID:263) #32 Elassona-TR 1 Yes (Node ID:269) #82 Sarantaporo-Makis 1 Yes (Node ID:283) #89 vaitsis-achilleas (Giannouli) 1 Yes (Node ID:265)

Figure 9.1: node installation in Tsapournia village

9.3.2. Social aspect

As explained in the Background Section, the Sarantaporo.gr CWN started as -and still is- a bottom-up initiative, meaning that people from local village societies are directly involved in the planning, deployment, maintainance and expansion of the CWN. The vast majority of the equipment was installed on locals’ rooftops, rendering these people participants to this endeavor. In order to incentivize more local people to participate more actively to the project we organized and implemented one info-point

Deliverable D.4.7 119 9.3. Test Setup and Results 9. Sarantaporo.gr

Figure 9.2: Tsapournia node RD installation

Figure 9.3: Sarantaporo.gr CONFINE testbed nodes

per village, where we presented an overview of the project and attempted to plan together with the locals the further use of this infrastructure for the development of the area. 12 info-points took place

Deliverable D.4.7 120 9. Sarantaporo.gr 9.4. Main achievements and challenges in total, for which a relative dissemination material was produced and distributed. 14 local support groups were created during these info-points, totalling 81 local people willing to contribute. An emailing list was set up as a communication channel with the local support groups. During the project we participated in three events, relative to our scope of work: • 2nd CommonsFest, Heraklion Crete, Greece, May 2014 • 3rd CommonsFest, Athens, Greece, May 2015 • Presentation of our project within the CONFINE framework in Hackerspace, Athens, Greece, July 2014 • Social and Solidarity Economy Festival, Athens, Greece, October 2014 Furthermore, we participated providing feedback in three papers on community networks (social and technical aspects).

9.4. Main achievements and challenges

Concerning the technical aspect of our project, the installation of the equipment was successfully completed on schedule and its operation was relatively smooth (with minor shortcomings that were timely dealt with). The infrastructure was used to create slivers for the experiments that run on the testbed. An overview of the Sarantaporo.gr sliver size relative to other participants’ is depicted in the following graph:

Figure 9.4: slivers allocation by node in the CONFINE Testbed

The results of the experiments that run on our infrastructure are discussed in detail in other sections of the project’s final report. In a social context the project also had an added-value effect for the local societies that was mani- fested through the building of an infrastructure as a common good. One of the major problems the local communities of the region face is the isolation from the urban centers, which is multiplied by the digital divide caused by lack of contemporary Internet connectivity. Telecommunication compa- nies (TelCos) are reluctant to invest in modern communication infrastructure in these areas due to minimum or non-existing profit margins. In cases where such infrastructure is provided it is often not affordable by locals, due to their insufficient income. This leaves local communities with the dilemma to be “disconnected” or to spend a considerable amount of their small income in order to keep up with the digital era (which is considered a human right in modern western societies).

Deliverable D.4.7 121 9.4. Main achievements and challenges 9. Sarantaporo.gr

Building our infrastructure in a bottom-up approach and making it available to everyone has had a disrupting effect on this situation. And this is not only due to the fact that local people have started embodying modern communication in their everyday lives. It’s also about realizing the potential to build and manage this infrastructure on their own and use it as a local development tool. During the past year we have witnessed multiple examples of how all this is transforming the local societies. Just to list a few: • More than 5.000 people are using the CWN for business, leisure, communication, knowledge- sharing, participation in governance and public administration, accessing public sector services etc. As an example, in July 2015, 5.5k devices exchanged a total of 5.5TB of data. • The villages’ visiting doctors are using the CWN infrastructure to access the online medicine prescription public service, in order to serve the local patients. • Visiting family members of the local elder residents have been observed to prolong their visiting duration in the villages during holidays, (long) weekends or even during weekdays, due to the availability of Internet connectivity. Working people are now able to work from a distance, while their kids can keep up with their friends over social networks (even being enthusiastic about sharing their village experiences). • Farmers of the area are using the web to reach out to new markets for their products and obtain better prices (some have been contacted by “no-middlemen” open-air market organizers), to find new raw material providers and discover cheaper, better-quality spare parts for their machinery. • Elders are using teleconferencing services to connect with their migrant children and grandchildren. As an elder woman told us: “we use to see our grandchildren every two to three years. Now we can see them growing, drinking our morning coffee together”! • Pupils are using Internet to socialize with their fellow-pupils and also to find online resources for the school projects. Some even praise their village to their fellow-students for having “open- access Internet connectivity”. • Local festivals and social activities can be live-streamed abroad, to members of the homogene- ity. • Strong WiFi signal spots are becoming a meeting point for locals, reviving in contemporary terms the custom of the “sycamore tree assembly” of the village. The dynamic created by the project also mobilized other local actors to participate, such as the Saran- taporo village church and a military camp located in the mountainside of Olympus. A CWN node has been installed in each of them, providing open access to local inhabitants, citizens serving their military service, army officers and passing-by hikers. Our CWN has been further utilized for research outside the CONFINE project by clinics of the Thes- saly University Medical School. Specifically, two questionnaires have been created and published through the CWN splash-screens to local inhabitants, asking for their feedback on the following sub- jects: • Chronic obstructive pulmonary disease • Sleep disorders The CWN infrastructure helped focusing the questionnaires to specific audience (villages’ inhabitants), which would otherwise have required cumbersome commute of the researches to the villages, travel and accommodation expenses, social interaction effort to convince locals to participate and a considerable amount of time to fulfill all these.

Deliverable D.4.7 122 9. Sarantaporo.gr 9.4. Main achievements and challenges

The above mentioned examples showcase the effect of the CWN in strengthening the social fabric and offering local development opportunities. They constitute inspiring examples for other communities, as we have witnessed in various events and meetings that we participated in the past one year and a half. A direct outcome of this has been the proposal we received in August 2014 by a documentarist to narrate the story of the Sarantaporo.gr CWN through a documentary. The work commenced in August 2014, but it is still a work-in-progress due to lack of budget. It is now in the editing stage and we expect it to be complete by the end of October 2015. In the course of the project we faced various challenges, some of which were evident at the beginning and some which became apparent as we progressed, both in technical and social aspects. Most of the technical challenges we faced were related to downtime of the nodes, primarily due to power outages. Although simple in its nature, this problem proved to be a challenge for our team due to lack of dedicated personnel, responsible for the equipment. The remote location of the villages was the reason that some of these problems were not promptly addressed. A significant challenge was the integration of our existing infrastructure (OpenMesh routers) with the new one (Mikrotik routers). Lacking a systematic benchmarking procedure, in order to identify bottlenecks, it was quite difficult to implement an optimal functionality of the CWN, which led to problems such as occasional slow connectivity or nodes disconnection (and also occasional grumbling from the CWN users. . . ). This was a problem caused mainly by the lacking of experts’ availability. We managed to successfully tackle these problems with the help of AWMN, a project partner with expertise and willingness to help. A challenge of a combined socio-technical nature, which emerged during the project, was associated with our choice of equipment, namely the Mikrotik routers. There has long been a debate among free/open-source communities about the feasibility of using proprietary instead of free software under certain circumstances. A bipole this discourse often boils down to is the ease-of-use vs free sharing of produced knowledge. Our decision to go for the proprietary alternative was mainly based on three facts: a) large distance of the project’s region from Athens, b) lack of local experts to support the infrastructure and c) the need to deliver the infrastructure on-time, as per the project schedule. For this decision of ours we received a rather strict criticism by members of Greek open-source communities, who argued that our project does not give back to the open-source community. Many attempts were made to work with these people on this issue, albeit ultimately unsuccessful. We tend to conclude that this is a discourse which needs to be addressed on a per-case basis, taking into consideration present strengths and weaknesses. The biggest challenge from social aspect that we came across was the active engagement of the local community to our endeavor. The difficulties we faced were rooted deeply in social automation evident in the local communities. The provided open WiFi connectivity in the villages was perceived from the locals as a “public service”, which “someone” was obliged to offer. In this sense their demand for reliable connectivity emerged in a very natural manner. That “someone” who was “obliged” to maintain the CWN was often assumed to be the local Municipality of Elassona, since not many were initially aware of the CONFINE project and the story behind Sarantaporo.gr CWN. A critical shortcoming from our side was the fact that we didn’t focus enough on building awareness around the bottom-up approach of our endeavor, but stressed the “free” aspect of the new Internet connectivity infrastructure instead, at least in the early stage of our work. Although we managed to create local support teams in almost every village, to undertake the basic maintenance of the infrastructure, we didn’t succeed in actively engaging these people. The main reasons behind that were the technical nature of the project, which discouraged the non-technically- savvy local volunteers, and the fact that we didn’t have members of our group permanently staying in

Deliverable D.4.7 123 9.5. Conclusions 9. Sarantaporo.gr the region, who would act as point of reference. The emailing list that we created in order to provide support to our local teams proved to be highly insufﬁcient. Direct communication via phone was much more successful in some cases. Last, but certainly not least, is the challenge we face with the sustainability of the project. The donations we receive currently are barely enough to cover for minimum expenses for maintenance and so far we haven’t been able to ensure permanent donors. For this to succeed we realize we need to work with the local communities in building awareness about the potential of the infrastructure to be used as a development tool; a tool that is managed by the community. To this end it is imperative that we focus on providing local, ﬁne-tuned services that address the needs and expectations of the end-users, such as telemedicine, distance education, precision agriculture etc.

9.5. Conclusions

Our participation in the CONFINE project offered us the opportunity to expand our CWN and interconnect the islet villages via a backbone infrastructure. It further enabled us to expand to the nearest Technical Institute and through this to other community networks all over Europe. Through this program we managed to increase the value of our infrastructure and open a contemporary communication channel for the local communities to utilize. At the same time we contributed to the experimental testbed of the CONFINE project by adding more research devices and offering greater diversity to the overall setup. Having gained significant experience and expertise in deploying and managing our infrastructure, we have also witnessed thus far how this infrastructure can provoke significant changes to the daily lives of local inhabitants. We have realized the great native development potential of this technology that can be harnessed for the benefit of the local communities. A lot of development opportunities are open in the perspective of connecting more villages of the region, even breaking the administrative barriers and expanding to neighboring regions as well. In view of the latest technological advances in Internet connectivity, with major providers attempting to provide world-wide WiFi coverage, we realize it is imperative to develop and provide new local services, fine-tuned for the needs of local inhabitants, which will be deployed on our infrastructure. This could be the cornerstone of a business model that would render the whole project sustainable, and spur the active engagement of local communities to the development, maintenance and expansion of the Sarantaporo.gr community wireless network.

Deliverable D.4.7 124 10. Conclusions

This deliverable reports on the ﬁnal results or the status of the second open call of the CONFINE project. The results show the various possibilities of the Community-Lab testbed, and include some recommendations to the CONFINE consortium. The experiments performed during the second open call can be divided into three categories: technological experiments, social or socioeconomical experiments and testbed expansion.

10.1. Technological Experiments

Four technological experiments have been carried out during the second open call: COSMOS, Min- strelBlues, Reflection and BTC. In COSMOS, the University of Cambridge and the Leibniz Universitat¨ Hannover investigated the benefits of extending the coverage of any crowd-shared network by connecting the home routers as a mesh. They showed that the advance knowledge of user sharing policies can reduce the number of flow redirections (particularly for low and moderate network loads) avoiding implications, such as packet reordering and communication overhead. By using an SDN-based controller to incorporate the sharing policies in a simple centralized algorithm, they are able to perform traffic redirection leading to a high utilization of the shared bandwidth, as opposed to a crowd-shared network with a single point of access where a significant amount of shared bandwidth is wasted. The Technische Universitat¨ Berlin implemented a cross-layer joint power and rate controller, Minstrel-Blues, which enables realising different power control approaches using today’s WiFi hardware. They have performed experiments that show that in both phases where Transmit Power Control is enabled, Minstrel-Blues does reduce its power level significantly. Therefore the overall interference to the environment is reduced. One of the main outcomes of the participation of Routek in the second open call in CONFINE, is the Network Characterisation Daemon (NCD). The NCD is conceived as a decentralised software running independently on each node in the community network. It retrieves and modifies status data and configuration from both the local and the other nodes via the NCD daemon. Additionally, it implements mechanisms to exchange information with other NCD daemons running on the other nodes of the community network. The opportunity to use the WiBed testbed has been of great help to test their software in a controlled environment very representative of the target site of typical network deployments developed by Routek. The Telecommunications Research Center Vienna participated in the open call with the BTC project (“Blessing of the Commons”). Their experiment can be divided in two main parts: the first part is an empirical model of interference in IEEE 802.11, where they present methodology, experiment config- uraton and final results from a set of measurements which provide detailed insight on the characteristics of current receiver implementations in terms of their capability to successfully decode OFDM frames which are overlapped in time by one or more lower-power frames received on the same carrie frequency. The second part covers resource utilization awareness in IEEE 802.11. There, they present some of the modifications to the Linux kernel driver code and selected user-space tools they have im-

125 10.2. Social Experiments 10. Conclusions plemented. The channel-load and spectral efﬁciency estimation methods described in this deliverable are continuously being tested and further validated using the 10-node OpenWRT/TP-WDR4300/ath9k indoor mesh that has been setup speciﬁcally for that purpose.

10.2. Social Experiments

Itinerarium and the Institute of Government and Public Policies of the Autonomous University of Barcelona participated in CONFINE with their CitizenSqKm project. It contributes to a better governance and policy making in new social practices of innovation; it is already being replicated locally, in three cities in the province of Barcelona and a possible pilot to be conducted in South Africa is being explored. From a methodological innovation perspective, CitizenSqKm has been an applied case of social innovation based on geolocated data, adding value to the state of the art in the social science. The project provided the tools and technologies for the local community to engage and collaborate, increasing the degree and quality of the knowledge citizens have about the area of the selected quarter, helping identify and solve common problems. The New America Foundation has made a contextual analysis of community network sustainability. Its contribution to the CONFINE project has been to develop a broad-based framework for understanding socioeconomic factors that determine the success, sustainability, and social impact of community networks; and to provide policymakers, as well as community networking advocates, researchers, and practitioners with a set of tools and best practices for building sustainability and positive impact into community networking practices. In their findings, they state that “as cities and towns work towards planning more collaborative, redundant, flexible, and ecologically adaptive systems in general, broadband infrastructure can be a site of pioneering cooperation. However, there are significant gaps in the field of community networking in availability of information about social impact and potential.” Project Icarus, by the University of the Western Cape, introduces the case of a community network in Mankosi, in rural South Africa and presents the results of a study to analyse the impact of connecting it to the Internet. Access to the Internet has been proven to have a major impact in the community with local students who managed to complete their application to a Higher Education Institution being granted a bursary to continue their education. The project has focused on studying four concrete aspects of the Community Network in Mankosi: • The impact of the communication network on the communication expenditure and patterns of both mobile and non-mobile users. • The spill over effects of the community network and the Internet connectivity. • The influence of the community network in the agency and aspirations of users. • The business model used to sustain the community network and its services.

10.3. Testbed Expansion

In CONFLATE (CONFINE extension towards OpenFlow experimentation: infrastructure, software and demonstrations), UNIDATA and the Consorzio Nazionale Interuniversitario per le Telecomuni- cazione have expanded the Community-Lab testbed in two dimensions - infrastructural and functional - by deploying new Community-Lab research devices in the Ninux.org network in Rome and by allowing researchers to deploy OpenFlow experiments in Community-Lab. With the OFX extension,

Deliverable D.4.7 126 10. Conclusions 10.3. Testbed Expansion

Community-Lab is, to our knowledge, the first wireless testbed which enables OpenFlow experiments within a large scale production network. To test the effectiveness of OFX, they have performed an OpenFlow-based experiment employing slivers inside the Ninux.org community network, in which they compare the performance of different Adaptive Bit Rate (ABR) strategies for the download of MPEG-DASH video segments. The goal of Sarantaporo.gr Non Profit Association’s participation in the CONFINE project under Open Call 2, was to provide expansion of the existing project’s testbed for enriching the experimental and testing environment. Being very specific with its rural character and a socio-economic status of underdeveloped region, with no other alternative for modern connectivity in almost half of the villages in the network, the Sarantaporo area was a suitable case to be added to the community lab testbed in terms of diversifying the characteristics under research of community wireless networks.

Deliverable D.4.7 127 Bibliography

[1] “IEEE std 802.11-2012. part 11: Wireless lan medium access control (mac) and physical layer (phy) specifications,” 2012. (document), 4.3.4, 4.11 [2] L. Townsend, A. Sathiaseelan, G. Fairhurst, and C. Wallace, “Enhanced broadband access as a solution to the social and economic problems of the rural digital divide,” Local Economy, vol. 28, no. 6, pp. 580–595, Sept. 2013. 1.1 [3] J. Fildes, “UN reveals global disparity in broadband access,” http://www.bbc.co.uk/ news/technology-11162656, 2010. 1.1 [4] A. Sathiaseelan et al., “Public access wifi service (PAWS),” Digital Economy All Hands Meet- ing, Aberdeen, october 2012. 1.1 [5] “Freifunk,” http://freifunk.net/. 1.1, 1.2.1 [6] A. Sathiaseelan, C. Rotsos, C.S. Sriram, D. Trossen, P. Papadimitriou, and J. Crowcroft, “Virtual public networks,” in Software Defined Networks (EWSDN), 2013 Second European Workshop on, Oct 2013, pp. 1–6. 1.2.1, 1.2.3 [7] B. Braem, C. Blondia, C. Barz, H. Rogge, F. Freitag, L. Navarro, J. Bonicioli, S. Papathanasiou, P. Escrich, R. Baig Vi nas, A.L. Kaplan, A. Neumann, I. Vilata, I. Balaguer, B. Tatum, and M. Matson, “A case for research with and on community networks,” SIGCOMM Comput. Commun. Rev, vol. 43, no. 3, pp. 68–73, July 2013. 1.2.1, 7.1 [8] B. Frank, I. Poese, Y. Lin, G. Smaragdakis, A. Feldmann, B. Maggs, J. Rake, S. Uhlig, and R. Weber, “Pushing CDN-ISP collaboration to the limit,” SIGCOMM Comput. Commun. Rev., vol. 43, no. 3, pp. 34–44, July 2013. 1.2.1 [9] S. Even, A. Itai, and A. Shamir, “On the complexity of time table and multi-commodity flow problems,” in Foundations of Computer Science, 1975., 16th Annual Symposium on, Oct 1975, pp. 184–193. 1.2.2 [10] Open Networking Foundation, “OpenFlow Switch Specification, version 1.3.3,” Sep 2013. 1.2.4.1, 8.1 [11] M. Jain and C. Dovrolis, “End-to-end available bandwidth: Measurement methodology, dynamics, and relation with tcp throughput,” IEEE/ACM Trans. Netw., vol. 11, no. 4, pp. 537–549, Aug. 2003. 1.2.4.1 [12] B. Melander, M. Bjorkman, and P. Gunningberg, “A new end-to-end probing and analysis method for estimating bandwidth bottlenecks,” in Global Telecommunications Conference, 2000. GLOBECOM ’00. IEEE, 2000, vol. 1, pp. 415–420 vol.1. 1.2.4.1 [13] Ningning Hu and P. Steenkiste, “Evaluation and characterization of available bandwidth probing techniques,” Selected Areas in Communications, IEEE Journal on, vol. 21, no. 6, pp. 879–894, Aug 2003. 1.2.4.1 [14] Eddie Kohler, Robert Morris, Benjie Chen, John Jannotti, and M Frans Kaashoek, “The click modular router,” ACM Transactions on Computer Systems (TOCS), vol. 18, no. 3, pp. 263–297, 2000. 1.2.4.2 [15] OPEN VSWITCH, “Open vswitch,” 2013. 1.2.4.2

128 Bibliography Bibliography

[16] “Multi-generator (MGEN),” http://www.nrl.navy.mil/itd/ncs/products/ mgen. 1.4 [17] N. Bambos, “Toward power-sensitive network architectures in wireless communications: concepts, issues, and design aspects,” Personal Communications, IEEE, vol. 5, no. 3, pp. 50 –59, June 1998. 2.2 [18] V. Kawadia and P.R. Kumar, “Principles and protocols for power control in wireless ad hoc networks,” Selected Areas in Communications, IEEE Journal on, vol. 23, no. 1, pp. 76 – 88, Jan. 2005. 2.2 [19] G.J. Foschini and Z. Miljanic, “A simple distributed autonomous power control algorithm and its convergence,” Vehicular Technology, IEEE Transactions on, vol. 42, no. 4, pp. 641 –646, Nov. 1993. 2.2 [20] Guangyu Pei and Thomas R. Henderson, “Validation of OFDM error rate model in ns-3,” Tech- nical note. http://www.nsnam.org/˜pei/80211ofdm.pdf, 2010. 4.2.2 [21] L. E. Miller, “Validation of 802.11a/uwb coexistence simulation,” Technical Report. http:// www.antd.nist.gov/wctg/manet/docs/coexvalid_031017.pdf, 2003. 4.2.2 [22] Jeongkeun Lee, Wonho Kim, Sung-Ju Lee, Daehyung Jo, Jiho Ryu, Taekyoung Kwon, and Yanghee Choi, “An experimental study on the capture effect in 802.11a networks,” in 2nd ACM International Workshop on Wireless Network Testbeds, Experimental Evaluation and Charac- terization (WinTECH’07). 2007, pp. 19–26, ACM. 4.2.2, 4.2.5 [23] A. Kochut, A. Vasan, A.U. Shankar, and A. Agrawala, “Snifﬁng out the correct physical layer capture model in 802.11b,” in 12th IEEE International Conference on Network Protocols (ICNP 2004), Oct 2004, pp. 252–261. 4.2.2 [24] Jiho Ryu, Jeongkeun Lee, Sung-Ju Lee, and Taekyoung Kwon, “Revamping the IEEE 802.11a PHY simulation models,” in 11th International Symposium on Modeling, Analysis and Simula- tion of Wireless and Mobile Systems (MSWiM ’08). 2008, pp. 28–36, ACM. 4.2.2 [25] Jeongkeun Lee, Jiho Ryu, Sung-Ju Lee, and Ted Taekyoung Kwon, “Improved modeling of IEEE 802.11a PHY through ﬁne-grained measurements,” Computer Networks, vol. 54, no. 4, pp. 641–657, 2010, Advances in Wireless and Mobile Networks. 4.2.2 [26] P. van Wijngaarden, “Frame capture in IEEE 802.11p vehicular networks: a simulation-based approach,” M.S. thesis, University of Twente, July 2011. 4.2.2 [27] Hoon Chang, V. Misra, and D. Rubenstein, “A general model and analysis of physical layer capture in 802.11 networks,” in 25th IEEE International Conference on Computer Communications (INFOCOM 2006), April 2006, pp. 1–12. 4.2.2 [28] C. Ware, J. Chicharo, and T. Wysocki, “Simulation of capture behaviour in IEEE 802.11 radio modems,” in IEEE VTS 54th Vehicular Technology Conference (VTC 2001 Fall), 2001, vol. 3, pp. 1393–1397 vol.3. 4.2.2 [29] Felix Schmidt-Eisenlohr, Interference in vehicle-to-vehicle communication networks - analysis, modeling, simulation and assessment, Ph.D. thesis, Karlsruhe Institute of Technology (KIT), 2010. 4.2.2 [30] Jens Mittag, Characterization, Avoidance and Repair of Packet Collisions in Inter-Vehicle Com- munication Networks, Ph.D. thesis, University of Karlsruhe, Germany, 2012. 4.2.2 [31] Mathieu Lacage and Thomas R. Henderson, “Yet another network simulator,” in Proceeding from the 2006 workshop on ns-2: the IP network simulator, 2006, WNS2 ’06. 4.2.3 [32] P. Fuxjager,¨ A. Costantini, D. Valerio, P. Castiglione, G. Zacheo, T. Zemen, and F. Ricciato,

Deliverable D.4.7 129 Bibliography Bibliography

“IEEE 802.11p Transmission Using GNURadio,” in 6th Karlsruhe Workshop on Software Ra- dios (WSR), 2010, pp. 1–4. 4.2.4.1, 4.2.4.2 [33] P. Gupta and P. R. Kumar, “The capacity of wireless networks,” IEEE Transactions on Infor- mation Theory, vol. 46, no. 2, pp. 388–404, 2000. 4.3.1 [34] Christoph Barz and Henning Rogge, “Improved community network node design using a dlep based radio-to-router interface,” in 1st International Workshop on Community Networks and Bottom-Up Broadband (CNBuB 2012), 2012. 4.3.1, 4.3.8 [35] Douglas S. J. De Couto, High-throughput Routing for Multi-hop Wireless Networks, Ph.D. thesis, Massachusetts Institute of Technology, Cambridge, MA, 2004. 4.3.2 [36] Vineeth Kisara, “A new routing metric for wireless meshes,” M.S. thesis, Iowa State University, 2010. 4.3.2 [37] Nadeem Javaid, Akmal Javaid, Imran Ali Khan, and Karim Djouani, “Performance study of ETX based wireless routing metrics,” in 2nd IEEE International Conference on Computer, Control and Communication (IEEE-IC4), 2009. 4.3.2 [38] Stefan Ruehrup Paul Fuxjaeger and Dieter Smely, “Impact of cca threshold, contention window, and transmit rate on vanet simulations,” in 39th Annual Conference of the IEEE Industrial Electronics Society (IECON 2013), 2013. 4.3.3 [39] “Archive of the ath9k-dev,” Mailing list. http://lists.ath9k.org/pipermail/ ath9k-devel/. 4.3.3 [40] “Acs (automatic channel selection) developer documentation,” Technical note. https:// wireless.wiki.kernel.org/en/users/documentation/acs. 4.3.3 [41] Burak Simsek, Katinka Wolter, and Hakan Coskun, “Analysis of the qbss load element parameters of 802.11e for a priori estimation of service quality,” International Journal of Simulation: Systems, Science and Technology, Special Issue: Performance Engineering of Computer and Communication Systems, 2006. 4.3.4 [42] “horst - lightweight wireless lan scanner and analyzer,” Code repository. http://br1. einfach.org/tech/horst/. 4.3.6 [43] Doug Schuler, “Community networks: building a new participatory medium,” Communications of the ACM, vol. 37, no. 1, pp. 38–51, 1994. 6.2 [44] Andrea H Tapia and Julio Angel Ortiz, “Keeping promises: Municipal communities struggle to fulﬁll promises to narrow the digital divide with municipal community wireless networks,” The Journal of Community Informatics, vol. 4, no. 1, 2008. 6.2 [45] Alison Powell, “Wiﬁ publics: Producing community and technology,” Information, Communi- cation & Society, vol. 11, no. 8, pp. 1068–1088, 2008. 6.4.2 [46] Maria Bina, “Wireless community networks: A case of modern collective action,” Athens University of Economics and Business, 2007. 6.4.2 [47] GreTA BYrum, “Building broadband commons,” New America., 2015. 6.4.2 [48] Juan Navas-Sabater, Andrew Dymond, and Niina Juntunen, Telecommunications and information services for the poor: Toward a strategy for universal access, Washington, DC: World Bank, 2002. 7.1 [49] J.C. Aker and I.M. Mbiti, “Mobile Phones and Economic Development in Africa,” Journal of Economic Perspectives, vol. 24, no. 3, pp. 207–232, 2010. 7.1 [50] Rudi Westerveld, “Inverse Telecommunications: The Future for Rural Areas in Developing

Deliverable D.4.7 130 Bibliography Bibliography

Countries?,” in Inverse Infrastructures: Disrupting Networks from Below, T M Egyedi and D C Mehos, Eds., pp. 187–207. Edward Elgar Publishing Ltd, 2012. 7.1 [51] Nicola J. Bidwell, M. Lalma, Gary Marsden, B. Dlutu, S. Ntlangano, A. Manjingolo, William D. Tucker, M. Jones, R. Robinson, E. Vartianen, and I. Klanpanos, “Please call ME.N.U.4EVER: Designing for ’Callback’ in Rural Africa,” in IWIPS’11 (10th Intl. Workshop on Internationali- sation of Products and Systems), 2011, pp. 117–138. 7.1, 7.4.1, 7.4.4 [52] David Johnson and Kobus Roux, “Building Rural Wireless Networks: Lessons Learnt and Future Directions,” in WiNS-DR’08, 2008, pp. 17–22. 7.1 [53] K.W Mathee, Gregory Mweemba, A.V. Pais, Gerjan Van Stam, and M. Rijken, “Bringing internet connectivity to rural zambia using a collaborative approach,” in International Conference on Information and Communication Technologies and Development (ICTD 2007), 2007, pp. 1–12. 7.1 [54] H Galperin and F Bar, “The microtelco opportunity: evidence from Latin America,” ITID, vol. 3, no. 2, pp. 73–86, 2006. 7.1 [55] S. O’Siochru and B. Girard, Community-based Networks and Innovative Technologies: New models to serve and empower the poor, United Nations Development Program, 2005. 7.1 [56] Eric Brewer, Michael Demmer, Melissa Ho, R J Honicky, Joyojeet Pal, Madelaine Plauche,´ and Sonesh Surana, “The Challenges of Technology Research for Developing Regions,” Pervasive Computing, vol. 5, no. 2, pp. 15–23, 2006. 7.1 [57] Hanna Hye-Na Cho, “Towards place-peer community and civic bandwidth: a case study in community wireless networking,” The Journal of Community Informatics, vol. 4, no. 1, 2008. 7.1 [58] Laura Forlano, “Anytime? anywhere?: Reframing debates around community and municipal wireless networking,” The Journal of Community Informatics, vol. 4, no. 1, 2008. 7.1 [59] Christian Sandvig, “An initial assessment of cooperative action in wi-ﬁ networking,” Telecom- munications Policy, vol. 28, no. 7, pp. 579–602, 2004. 7.1 [60] Andrea L Kavanaugh and Scott J Patterson, “The impact of community computer networks on social capital and community involvement,” American Behavioral Scientist, vol. 45, no. 3, pp. 496–509, 2001. 7.1 [61] J. Tacchi, D. Slater, and G. Hearn, Ethnographic Action Research. A user’s handbook developed to innovate and research ICT applications for poverty eradication, UNESCO, 2003. 7.2 [62] C. Rey-Moreno, Z. Roro, W.D. Tucker, M.J. Siya, N.J Bidwell, and J. Simo-Reigadas, “Ex- periences, challenges and lessons from rolling out a rural WiFi mesh network,” in 3rd ACM Symposium on Computing for Development. Mar 2013, ACM. 7.2 [63] Monde B. Makiwane and Dan O.D. Chimere-Dan, The People Matter: The State of the Popu- lation in the Eastern Cape, Research and Population Unit. Eastern Cape Department of Social Development, 2010. 7.2 [64] N. J. Bidwell, M. Siya, G. Marsden, W. D. Tucker, M. Tshemese, N.Gaven, S. Ntlangano, K. A. Eglinton, and S. Robinson, “Walking and the Social Life of Solar Charging in Rural Africa,” ACM Transactions on Computer-Human Interaction, vol. 20, no. 4, 2013. 7.2, 7.4.1, 7.4.4 [65] Carlos Rey-Moreno, Zukile Roro, Masbulele J. Siya, Javier Simo-Reigadas, Nicola J. Bidwell, and William D. Tucker, “Towards a Sustainable Business Model for Rural Telephony,” in Proc. of the III International Workshop on Research on ICT for Human Development, 2012. 7.2 [66] C. Rey-Moreno, W.D. Tucker, D. Cull, and R. Bloom, “Making a Community Network Legal

Deliverable D.4.7 131 Bibliography Bibliography

within the South African Regulatory Framework,” in Proceedings of the 7th International Con- ference on Information and Communications Technologies and Development: Notes. 2015, vol. forthcoming, ACM. 7.2, 7.4.4 [67] Carlos Rey-Moreno, Amalia Sabiescu, and Masbulele Siya, “Towards self-sustaining community networks in rural areas of developing countries: understanding local ownership,” in Pro- ceedings of the 8th International Development Informatics Association Conference, 2014, pp. 63–77.2 [68] C. Rey-Moreno, R. Blignaut, W.D. Tucker, and J. May, “An in-depth study of the ICT ecosystem in a South African rural community: unveiling expenditure and communication patterns,” Journal of Information Technology for Development, vol. In Press, 2015. 7.4.1, 7.4.2, 7.4.4 [69] M. Oliver, J. Zuidweg, and M. Batikas, “Wireless commons against the digital divide,” in IEEE International Symposium on Technology and Society (ISTAS), 2010, pp. 457–465. 7.4.2 [70] C. Rey-Moreno, A. Sabiescu, and W.D. Tucker, “Local ownership, exercise of ownership and moving from passive to active entitlement: A practice-led inquiry on a rural community network,” Journal of Community Informatics, vol. 11, no. 2, 2015. 7.4.2 [71] C. Rey-Moreno, M.J. Ufitamahoro, I. Venter, and W.D. Tucker, “Co-designing a Billing System for Voice Services in Rural South Africa: Lessons Learned,” in ACM DEV-5, 2014. 7.4.4 [72] DTI, “Co-operative Incentive Scheme (CIS),” Last accessed 14th February 2015. https://www.thedti.gov.za/financial assistance/financial incentive.jsp?id=11&subthemeid=3, 2015. 7.4.4 [73] “Ninux roma routing architecture document version 1,” http://wiki.ninux. org/NetworkArchitecture?action=AttachFile&do=view&target= NinuxRoma-RoutingArchitecture-DocumentVersion1.pdf. 8.2.1 [74] “CONFINE research device HOWTO (in italian),” http://wiki.ninux.org/ ConfineNodeHowTo. 8.2.1, 8.5 [75] “How to create virtual machines for experiments and experimental services in CONFINE (in italian),” http://wiki.ninux.org/ConfineVMHowTo. 8.2.1, 8.5 [76] Mahalingam, Dutt, and Duda et al., “Virtual extensible local area network (VXLAN): A framework for overlaying virtualized layer 2 networks over layer 3 networks,” RFC 7348, IETF, 2014. 8.2.2 [77] “OFX source code repository,” http://redmine.confine-project.eu/ projects/ofx/repository. 8.2.2, 8.2.2.1, 8.2.2.1, 8.2.2.2, 8.4.1, 8.5 [78] “Paramiko, a python implementation of SSHv2,” http://www.paramiko.org/. 8.2.2.1 [79] Ryu project team, “Ryu SDN framework,” Jun 2015, http://osrg.github.io/ ryu-book/en/Ryubook.pdf. 8.2.2.1 [80] “DASH Industry Forum,” http://dashif.org/. 8.3 [81] Danny H Lee, Constantine Dovrolis, and Ali C Begen, “Caching in http adaptive streaming: Friend or foe?,” in Proceedings of Network and Operating System Support on Digital Audio and Video Workshop. ACM, 2014, p. 31. 8.3 [82] Christopher Mueller, Stefan Lederer, and Christian Timmerer, “A proxy effect analyis and fair adatpation algorithm for multiple competing dynamic adaptive streaming over http clients,” in Visual Communications and Image Processing (VCIP), 2012 IEEE. IEEE, 2012, pp. 1–6. 8.3 [83] “dash.js reference client implementation,” http://dashif.org/reference/ players/javascript. 8.3

Deliverable D.4.7 132 Bibliography Bibliography

[84] “The apache HTTP server project,” http://www.apache.org/. 8.4.1 [85] “Apache tomcat,” https://tomcat.apache.org/. 8.4.1 [86] “squid : Optimising web delivery,” http://www.squid-cache.org/. 8.4.1 [87] Blender Institute, “Big buck bunny,” https://peach.blender.org/. 8.4.1

Deliverable D.4.7 133 The CONFINE project

September 2015

CONFINE-201510-D.4.7-1.0

This work is licensed under a Creative Commons “Attribution-ShareAlike 3.0 Unported ” license.