Plutarch: An Argument for Network PluralismJon CrowcroftUniversity of Cambridge15 JJ Thomson AvenueCambridge CB3 0FD, [email protected] HandUniversity of Cambridge15 JJ Thomson AvenueCambridge CB3 0FD, [email protected] MortierMicrosoft Research7 JJ Thomson AvenueCambridge CB3 0FB, [email protected] RoscoeIntel Research2150 Shattuck AveBerkeley, CA 94704, [email protected] WarfieldUniversity of Cambridge15 JJ Thomson AvenueCambridge CB3 0FD, [email protected] is widely accepted that the current Internet architectureis insufficient for the future: problems such as address spacescarcity, mobility and non-universal connectivity are alreadywith us, and stand to be exacerbated by the explosion ofwireless, ad-hoc and sensor networks. Furthermore, it is farfrom clear that the ubiquitous use of standard transport andname resolution protocols will remain practicable or evendesirable.In this paper we propose Plutarch, a new inter-networkingarchitecture. It subsumes existing architectures such as thatdetermined by the Internet Protocol suite, but makes ex-plicit the heterogeneity that contemporary inter-networkingschemes attempt to mask. To handle this heterogeneity,we introduce the notions of context and interstitial func-tion, and describe a supporting architecture. We discussthe benefits, present some potential scenarios, and considerthe research challenges posed.1. INTRODUCTIONThe remarkable success of the global Internet is frequentlyattributed to a set of design decisions that prioritize sim-plicity and robustness through a strongly specified suite ofprotocols, and the incorporation of only essential mecha-nisms within the network itself. The astounding growth ofthis network over the past thirty years serves as a clear tes-tament to the wisdom of these principles.However, in recent years, the Internet’s architectural as-sumptions have been fundamentally challenged. In particu-lar, the introduction of specialized networks such as sensornetworks, along with various middleboxes have all begunto strain the existing framework. As a result, we proposePermission to make digital or hard copies of all or part of this work forpersonal or classroom use is granted without fee provided that copies arenot made or distributed for profit or commercial advantage and that copiesbear this notice and the full citation on the first page. To copy otherwise, torepublish, to post on servers or to redistribute to lists, requires prior specificpermission and/or a fee.ACM SIGCOMM 2003 Workshops August 25&27, 2003, Karlsruhe, Ger-manyCopyright2003ACM1-58113-748-6/03/0008...$5.00.Plutarch, a new framework for next generation networks.It differs from the existing Internet architecture primarilyin that it embraces heterogeneity in the hope of allowingradical innovation. The homogeneous Internet architectureand its advantages are not abandoned but retained as onearchitecture among many.In Plutarch we divide the world into contexts, each compris-ing some set of hosts, routers, switches, network links and soforth. Within a context we expect homogeneity regardingsuch things as addresses, packet formats, transport proto-cols and naming services. Distinct contexts differ in at leastone of these areas.Communication across a set of contexts is enabled by in-terstitial functions, which map between the sets of func-tionalities encapsulated by contexts. We can divide suchfunctionalities into four main areas:Addressing: Mapping between different address contextsis a well-understood problem, and one that exists today inNAT boxes, for example. We suggest that programmaticinterfaces to such functionality should be exposed, allow-ing mappings to be automatically set up, maintained andmanaged by network users.Naming: Today, DNS provides a single global namespace,with management handled by hierarchical delegation. Wepredict that emerging services such as VoIP and personalarea networking will favour an alternative approach of map-ping between a plurality of naming systems, for reasons ofscalability and administrative overhead.Routing: Different styles of routing protocol are appropri-ate in different networks. For example, connecting an ad-hocwireless network with on-demand routing to an Internet ASrunning OSPF and BGP requires a more complex mappingthan simply exposing the routes of each network via BGP.Using an explicit interstitial function to map routing infor-mation between these networks hides the churn of Internetroutes from the on-demand protocol, deals with the wirelessnetwork protocol’s broadcast, and hides the instability andmutability of routes within the wireless network.Proceedings ot the ACM SIGCOMM 2003 Workshops 258 August 2003Transport: A single transport protocol struggles to dealwith all network technologies; symptomatic of this prob-lem is the example of wireless-specific TCP implementationsthat use techniques such as splitting and proxying (e.g. [9]).Optimizing transport protocols for specific network typeshas many advantages (for example, the use of large fat pipesin Grid infrastructures) but these come at the cost of work-ing out when such optimizations are appropriate. Explicitinterstitial functions provide well-defined points at whichsuch decisions can be taken.By not limiting the set of interstitial functions that may bedefined and built we hope to support extensibility within ex-isting networks, and encourage innovation within new ones.2. MOTIVATIONWe present the Plutarch framework for a number of rea-sons. The principal one is the concrete problem of con-necting networks where a common overlay protocol such asIPv4 or IPv6 is infeasible or undesirable, for example sensornetworks, or specialized networks which offer valuable intra-domain functionality which IP must ignore. We discuss thisissue, and how our solution addresses it, in the rest of thispaper.The second reason is that the abstract model underlyingPlutarch captures the state of the Internet we see todaybetter than models based on the Internet’s original archi-tectural principles [11]. This is not to say these principlesare wrong or inappropriate; rather, we claim that a concep-tual framework based upon them does not facilitate clearthinking about the future of the network. They may be theright principles, but they are not an adequate frame.Finally, a model of networking based on explicit contextsprovides a clearer framework within which to debate
View Full Document
Unlocking...