Unmanaged Internet ProtocolTaming the Edge Network Management CrisisBryan FordMassachusetts Institute of TechnologyAbstractThough appropriate for core Internet infrastructure, theInternet Protocol is unsuited to routing within and be-tween emerging ad-hoc edge networks due to its de-pendence on hierarchical, administratively assigned ad-dresses. Existing ad-hoc routing protocols address themanagement problem but do not scale to Internet-widenetworks. The promise of ubiquitous network computingcannot be fulfilled until we develop an Unmanaged Inter-net Protocol (UIP), a scalable routing protocol that man-ages itself automatically. UIP must route within and be-tween constantlychanging edge networks potentially con-taining millions or billions of nodes, and must still func-tion within edge networks disconnected from the mainInternet, all without imposing the administrative burdenof hierarchical address assignment. Such a protocol ap-pears challenging but feasible. We propose an architecturebased on self-certifying, cryptographicnode identities anda routing algorithm adapted from distributed hash tables.1 IntroductionThe promise of ubiquitous computing is that people willsoon routinely own many “smart” networked devices,some mobile, others perhaps built into their homes andoffices, and all of which they can access and control fromany location so long as appropriate security precautionsare taken. Before we can expect ordinary, non-technicalpeople to adopt this vision, however, the ad-hoc edge net-works in which these devices live must be able to managethemselves. Each device must be able to find and com-municate with its peers—whether connected directly, in-directly over a local-area network, or remotely across along distance via the Internet—with no special configura-tion or other technical effort on the part of the user.The current Internet Protocol is unsuited to this pur-pose. IPv4 and IPv6, with their accompanying routing,naming, and management protocols, have evolved aroundthe requirements of core network infrastructure: corpo-rate, academic, and government networks deployed andFigure 1: Today’s Internetworking Challengesmanaged by skilled network administrators. IP’s hierar-chical address architecture in particular is fundamentallydependent on skilled network management. Current ad-hoc networking protocols by themselves are not sufficienteither, because they are only scalable to local-area net-work sizes of a few hundreds or thousands of nodes.To achieve ubiquitous network computing, we need anUnmanaged Internet Protocol, or UIP, that combines theself-management of ad-hoc networks with the scalabilityof IP. As illustrated in Figure 1, achieving this goal in to-day’s chaotic mix of networking technologies also meansrouting traffic automatically and securely through NATs,and transparently bridging IPv4, IPv6, and other addressdomains. We propose an architecture based on scal-able identity-based routing, or routing based on topology-independent node identifiers. While more difficult thanrouting over topology-dependent addresses such as IPaddresses, there is evidence that scalable identity-basedrouting is possible and practical.This position paper is organized as follows. Section 2lays out the motivation for UIP and the inadequacies ofcurrent solutions. Section 3 proposes and outlines anidentity-based UIP routing architecture, and Section 4 de-scribes implementation status and deployment. Section 5summarizes related work, and Section 6 concludes.CCR Paper Comment, Discussion, and Update Forum: http://www.acm.org/sigcomm/CCR/forumACM SIGCOMM Computer Communications Review Volume 34, Number 1: January 2004932 Motivation for UIPThe original ARPAnet vision was to enable computerusers to communicate and share resources with users ofany other connected computer [18, 28]. This vision hasevolved into the modern Internet Protocol, whose purposeis to implement any-to-any connectivity between hosts,whether connected directly or indirectly via paths cross-ing many administrative domains. While physical andlink-layer technologies such as Ethernet providelow-levelbuilding blocks for communication, and higher-level pro-tocols enable applications and users to take advantage ofthe network, interoperable end-to-end connectivity via IPremains the Internet’s central focus.Technical, social, and economic pressures have hin-dered the achievement of this vision, however. The proto-cols underlying the Internet were designed by technicallysavvy individuals who understand how networks work butoften do not understand how non-technical users work.As a result many aspects of network operation still requirecareful and skilled management. We desire and increas-ingly expect that everyone should be able to use the Inter-net and deploy networked devices, and strong economicincentives exist for businesses to sell Internet-enabledhardware and Internet-based services to technically un-sophisticated users. Since businesses seek lowest-costpaths to profitable solutions, the commercial Internet hasevolved—via a chaotic series of hacks and extensions—into a system geared toward particular usage patterns thatfacilitate business opportunities, often at the expense ofinteroperability and general end-to-end connectivity.2.1 The Edge Network Management CrisisAn unsophisticated user can now buy a computer, connectit to the Internet, and use it for browsing the Web, readingE-mail, and shopping on-line. Users who are a bit moreadventurous but still relatively non-technical may set up asmall home network and surf the Web from several com-puters at once. But consider the following scenario:1. Joe User is working at home on his laptop. Hehas remote shell and database access sessions open,through his WiFi home network, to his desktop PCand to a machine at his workplace.2. Joe’s friend Jim calls and invites him over. Joe putshis laptop into sleep mode and hops into his car.3. Joe stops for a bite to eat on the way to Jim’s, andscribbles some notes on his PDA in the restaurant.4. Upon returning to his car, Joe tries to synchronize hisPDA with his laptop, but discovers they won’t talkto each other even though they’re both WiFi-enabledand are at most a foot apart. Being unfamiliar withthe technical details of IP networks, he doesn’t real-ize that this is because (a) the WiFi adapters are ininfrastructure rather than ad-hoc mode, and (b) evenif they could communicate at the link layer, neithermachine would
View Full Document
Unlocking...