Designed for Change: End-to-End Arguments, Internet Innovation, and the Net Neutrality Debate Richard Bennett Information Technology and Innovation Foundation Designed for Change: End-to-End Arguments, Internet Innovation, and the Net Neutrality Debate Richard Bennett September 2009 THE INFORMAtiON TECHNOLOGY & INNOVAtiON FOUNDAtiON Table of Contents I. Executive Summary ............................................................................................ 1 II. Introduction ......................................................................................................... 6 End-to-End in Network Engineering ............................................................... 7 A. ARPANET ............................................................................................... 7 B. CYCLADES ............................................................................................. 9 1. Flexibility ........................................................................................... 9 2. Simplicity .........................................................................................10 3. Multiple Gates ................................................................................10 4. The Influence of CYCLADES on Subsequent Networks ...... 11 C. The Connectionless vs. Connection-Oriented Debate ....................12 D. Moving From Practice to Theory .......................................................12 E. Single Points of Failure ........................................................................14 F. Internet Congestion Collapse ..............................................................15 The Rehabilitation of the End-to-End Arguments ......................................18 A. End-to-End as Limited Fate-Sharing .................................................18 B. Who Do You Trust? ..............................................................................19 C. Layering Considered Harmful .............................................................20 D. The Evolution of End-to-End ...........................................................20 E. Clark’s Tussle ..........................................................................................21 F. Current Status of End-to-End ............................................................22 End-to-End in Regulation ................................................................................23 A. From the Fringe to the Mainstream ...................................................26 B. Network Neutrality: The Birth of a Notion ......................................28 C. Functional Layering as a Network Regulatory Model ......................30 THE INFORMATION TECHNOLOGY & INNOVATION FOUNDATION | SEPTEMBER 2009 PAGE II Birth of the Public Internet ..............................................................................31 A. The End-to-End Tax: “World Wide Wait” .......................................33 B. Internet Innovation Cycle.....................................................................34 Challenges to the Internet Architecture ..........................................................34 A. Denial of Service Attacks ....................................................................34 B. Traffic Jams .............................................................................................34 C. Address Unknown .................................................................................35 D. The Proliferation of Prefixes ..............................................................35 E. LTE Needs Quality of Service for Voice ..........................................36 F. InterStream Needs Quality of Service for Video ............................36 G. RINA Rethinks Protocol Layering .....................................................36 Lessons from End-to-End’s Evolution ...........................................................37 A. In Search of a Regulatory Framework ...............................................38 B. Striking the Balance ..............................................................................39 Conclusion ...........................................................................................................39 Endnotes ..............................................................................................................41 List of Figures Figure 1: ARPANET Architecture ............................................................................. 8 Figure 2: CYCLADES Architecture........................................................................... 9 Figure 3: Internet Architecture .................................................................................11 Figure 4: Equal Pipes Congestion Scenario ............................................................17 Figure 5: Creating Congestion by Over-Provisioning. ..........................................25 THE INFORMATION TECHNOLOGY & INNOVATION FOUNDATION | SEPTEMBER 2009 PAGE III THE INFORMAtiON TECHNOLOGY & INNOVAtiON FOUNDAtiON Executive Summary dvocates of net neutrality stress the fact that the Internet’s The principle of constant free and open character stimulates innovation. This is un- change is perhaps the deniably true, of course: The fact that practically anyone only principle of the A with an idea can write an Internet program, test it, and deploy it on a Internet that should standard, world-wide platform at minimal expense is an essential part survive indefinitely. of the Internet’s appeal and the major generator of novel, useful and —RFC 1958 entertaining applications. Some neutralists argue that the Inter- coined the term “network neutrality,” net’s open character is the consequence concedes this point: 1 of a single design decision made more than 35 years ago involving the end-to- Proponents of open access have end principle, which in their view for- generally overlooked the fact that, bids network operators from actively to the extent an open access rule managing network traffic, even to im- inhibits vertical relationships, it can prove the consumer experience. What help maintain the Internet’s greatest this report shows, however, is that while deviation from network neutrality. end-to-end arguments reflect important That deviation is favoritism of data aspects of the Internet’s organization, applications, as a class, over latency- the belief that the Internet is built on sensitive applications involving voice negative principles—prohibitions on or video. taking action—is a serious misunder- standing of how the Internet was de- As new generations of application de- signed and how it works. signers imagine novel ways to employ networks, traditional paradigms of net- The Internet’s capacity to stimulate in- work usage can present obstacles—dis- novative new applications flows from its crimination by design—that can only ability to meet the needs of applications be overcome by improvements to the previously unimagined, not from delib- network in the form of service model erate impairment of the network core. modifications appropriate to new re- No single-service network is neutral, as quirements. The Internet was designed any given network service has greater for change: The real power of the end- utility for some types of applications to-end architecture lies in its ability than others. Tim Wu, the man who to stimulate innovation by providing service model enhancements that in turn enable new The commercial Internet has developed as a loose fed- types of applications at the network edge. Modern eration of separately owned and operated networks that networks are no longer constrained to provide one and have agreed to interconnect at a few hundred peering only one delivery service; they’re now capable of pro- points around the world on terms mutually agreed by viding multiple services, each tailored to a broad class each pair of networks. Internet applications running of applications, and allowing application developers to on end systems rely on this infrastructure to commu- choose the one that best meets their needs. nicate with each other. As the underlying technology improves, interconnection agreements occasionally The Internet’s flexibility comes from the decision to change as well; when such changes increase overall net- use a novel method of information transfer called the work capability, users and application providers stand “datagram” instead of a traditional telecom control sys- to benefit. Steady improvement, punctuated by brief tem. Datagram networks such as the Internet export periods of radical disruption, is the normal state of af- a significant portion of the network’s control logic to fairs in evolving technical systems and should not be end-user systems located on the network edge, where feared. It is likely that the only principle of the Internet it’s easy to experiment with policies and protocols sim- that should survive indefinitely is the principle of con- ply by writing new software; they are therefore termed stant change.2 “end-to-end” networks. End-to-end arguments de- scribe the structure of datagram networks and provide To understand the current debate over Internet regu- a blueprint for their continual improvement. lation, it’s necessary to appreciate that the end-to-end arguments of network engineering differ significantly Since the dawn of the datagram era, experimentation from network neutrality advocates’ idiosyncratic end- has identified communication functions that provide to-end principle, a demand for a low-function, “stupid” greatest utility