Show simple item record

Files in this item

Cover for Leveraging Network Maps to Improve Evaluations of Overlay System Performance and Security
dc.contributor.advisorSherr, Micahen
dc.creatoren
dc.date.accessioned2014-01-09T15:13:34Zen
dc.date.available2014-01-09T15:13:34Zen
dc.date.created2013en
dc.date.issueden
dc.date.submitted01/01/2013en
dc.identifier.otherAPT-BAG: georgetown.edu.10822_707416.tar;APT-ETAG: 71151e1f7e75928005cc67a566ec9b1fen
dc.identifier.urien
dc.descriptionM.S.en
dc.description.abstractDistributed systems often communicate through overlayen
dc.description.abstractnetworks, which use custom addressing and protocols toen
dc.description.abstractcommunicate between participating nodes at the application layer,en
dc.description.abstractbut route those custom messages over the standard networken
dc.description.abstractinfrastructure. Overlay networks enable application systemen
dc.description.abstractdesigners to focus on the intended operation of their systemen
dc.description.abstractdistinct from the network layer. This can have several benefits:en
dc.description.abstractimprovements at lower levels of the technology stack canen
dc.description.abstractbe assimilated without modifying the application layer protocol,en
dc.description.abstractand modeling application behavior is easier because the protocol doesn'ten
dc.description.abstractdepend on network interactions.en
dc.description.abstractHowever, while this enables overlay networks to be easily studieden
dc.description.abstractand modeled, this very abstraction can make it difficult toen
dc.description.abstractunderstand how their interaction with the network underlayen
dc.description.abstractaffects them. Ideally the application's behavior would been
dc.description.abstractcompletely isolated from the network layer, but in practice thisen
dc.description.abstractis rarely the case. For instance, application layer modelingen
dc.description.abstractcannot easily predict exactly how a widespread deployment willen
dc.description.abstractbehave; security and performance can both be affected by the pathen
dc.description.abstractoverlay networks take through the underlay network. This canen
dc.description.abstractmake conscientious operators of overlay networks hesistant toen
dc.description.abstractmake large modifications to their protocol for fear itsen
dc.description.abstractinteraction with lower layers, once distributed across the internet, will haveen
dc.description.abstractunintended effects. For instance, the Tor Project, which managesen
dc.description.abstractthe Tor anonymity network, is relativelyen
dc.description.abstractconservative with respect to protocol changes, in part because ofen
dc.description.abstractfears that a change might affect anonymity through someen
dc.description.abstractunexpected interaction with the underlying network, whether dueen
dc.description.abstractto routing or performance.en
dc.description.abstractThe goal of this thesis is to introduce network maps which canen
dc.description.abstractbe used to effectively evaluate overlay network technologies withen
dc.description.abstractrespect to both performance and anonymity within evaluationen
dc.description.abstractplatforms that provide a safe environment for experimentation.en
dc.description.abstractSafe evaluation environments are critical in that they permiten
dc.description.abstractmodification of core protocols without affecting active systemen
dc.description.abstractusers. We discuss the advantages and disadvantages posed byen
dc.description.abstractdifferent classes of evaluation platforms anden
dc.description.abstracthow they can interface with our proposed network maps.en
dc.description.abstractWe present a series of techniques for constructing these network maps which combine network information fromen
dc.description.abstractdisparate sources into large graphs which represent the globalen
dc.description.abstractinternet. For each type of network data, we discuss the sources from whichen
dc.description.abstractthey can be obtained and the types of inaccuracies they canen
dc.description.abstractintroduce in network evaluations. Given the set of availableen
dc.description.abstractdata, we propose methods for constructing network maps byen
dc.description.abstractcombining these sources of information.en
dc.description.abstractWe develop maps at two granularity levels, then present severalen
dc.description.abstractcase studies which use the proposed mapping techniques inen
dc.description.abstractcombination with several platforms to perform security anden
dc.description.abstractperformance evaluations of the Tor anonymity network, including aen
dc.description.abstractconsideration of the effects of modifications to the Toren
dc.description.abstractprotocol. The first study investigates the performance anden
dc.description.abstractsecurity implications of a number of modifications to Tor's relayen
dc.description.abstractselection strategy. We show that while Tor's existing strategy isen
dc.description.abstracthighly effective, there are opportunities for performanceen
dc.description.abstractimprovement from layered selection strategies. A second studyen
dc.description.abstractresearches the level and prevalence of the threat posed to Toren
dc.description.abstractusers by network level adversaries, showing that Tor users areen
dc.description.abstracthighly vulnerable -- perhaps more so than previously thought --en
dc.description.abstractagainst network adversaries.en
dc.formatPDFen
dc.format.extent91 leavesen
dc.languageenen
dc.publisherGeorgetown Universityen
dc.sourceGeorgetown University-Graduate School of Arts & Sciencesen
dc.sourceComputer Scienceen
dc.subjectmodelingen
dc.subjectnetwork evaluationen
dc.subjectoverlay systemen
dc.subject.lcshComputer scienceen
dc.subject.otherComputer scienceen
dc.titleLeveraging Network Maps to Improve Evaluations of Overlay System Performance and Securityen
dc.typethesisen


This item appears in the following Collection(s)

Show simple item record