/
Bo ycotting and Extor ting Nodes in an Internetw ork D Bo ycotting and Extor ting Nodes in an Internetw ork D

Bo ycotting and Extor ting Nodes in an Internetw ork D - PDF document

trish-goza
trish-goza . @trish-goza
Follow
444 views
Uploaded On 2015-06-15

Bo ycotting and Extor ting Nodes in an Internetw ork D - PPT Presentation

umdedu 1 INTR ODUCTION oyc ott is protest or demand for hange based on consumers follo wing simple rule do not purc hase service or go ds from sp eci57356c pro ducers In an net ork and esp ecially in the In ternet suc grassro ots protest is close to ID: 86461

umdedu INTR ODUCTION oyc

Share:

Link:

Embed:

Download Presentation from below link

Download Pdf The PPT/PDF document "Bo ycotting and Extor ting Nodes in an I..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.


Presentation Transcript

BoycottingandExtortingNodesinanInternetworkDaveLevinAdamBenderCristianLumezanuNeilSpringBobbyBhattacharjeeUniversityofMaryland,CollegeParkEmails:fdml,bender,lume,nspring,bobbyg@cs.umd.edu1.INTRODUCTIONAboycottisaprotestorademandforchangebasedonconsumersfollowingasimplerule:donotpurchaseserviceorgoodsfromspeci cproducers.Inanynetwork,andespeciallyintheInternet,suchgrass-rootsprotestisclosetoimpossible.Thebignetworksdonotconnectdirectlytoconsumersandroutingisbasedonlocallychoosingtheshortest,cheapest,ormostpro tablepaths.Inthisenvironment,individualconsumershavenoin\ruenceonrouteselectionandenterprisecustomersverylittle.Yet,protest-worthyactionsbynetworkprovidersarenotinfrequent.AOLcompromisedtheprivacyofitsusersbypublishingsearchqueries(August2006);Googlecom-pliedwithChina'scensorshiplaws,tothedisapprovalofmany(February2006);VerisignredirectedmistypedDNSnamestotheirownadvertising(September2003);andtheElectronicFrontierFoundationhassuedAT&TtostopNSAsurveillance(January2006;legalactioncontinues).WeenvisionanInternetinwhichusershavetheabilitytomake,oratleastin\ruence,routingdecisions.Userswillhavethisabilitysothattheymaytakeadvantageofincreasinglyrichfunctionalityinnetworks;userswillneedtobeabletochoosenetworksthatprovideaserviceoravoidthosethat lterorcensor.Amongthemanypoliciesusersmightchooseinroutingpackets,usersmayselectnodestoboycott.Tostudytheeconomicsofboycottingaproviderinanabstractnetwork,weadapttheVickrey-Clarke-Groves(VCG)routingmechanismtosupportboycotting.Alllinkcostsaregloballyadvertised.Sourcesinthisnetworktypicallychoosetheshortestcostpathtoadestination,andmicropaymentsrewardeachentityalongthepath.EachoftheseassumptionsisdistinctfromtherealitiesofBGP;wedescribetheimplicationsofthismismatchlateinthispaper.Webegintheinvestigationintothefollowingquestions:(1)DoesVCGcontinuetoencouragenodestoreportcoststruthfullyinthepresenceofboycotting?(2)Donodesgainfromlearningusers'boycottlists?Dousersgainfromdivulgingwhomtheyareboycotting?and(3)HowdotheresultsfromVCGapplytoadhocnetworksandtheInternet?Toanswerthesequestions,wemodifyVCG(x2)andshowthatonlywhenmassivelyboycotteddonodeshaveincentivetolieaboutlinkcosts(x3).Wesimulateourmodi edVCG(x4)toshowthepriceinconnectivity,anddiscussdeploymentrequirementsandimplications(x5).Wereviewrelatedwork(x6)andconclude(x7).2.AMECHANISMFORBOYCOTTING2.1Model,Assumptions,andProblemWemodelthenetworktopologyasadirected,weightedgraphG=(V;E).Foranytwonodesu;v2V,c(u;v)isthenonnegativecostincurredbyutosendonthedi-rectededge(u;v)2E.Anodecanhaveatleastoneofthefollowingroles:source,destination,ortransit.Eachsourcenodeshasa(possiblyempty)setBsofnodesthatitwishestoboycott.Thatis,sprefersthatnotracsentbysfollowsapaththatcontainsanyb2Bs.Weassumethatsourceroutingissupported,butshownextthatsourceroutingaloneisnotenoughtoboycott.Weassumeapaymentmechanismbywhichscansecurelypayothernodesinthenetwork,andanaccountabilitymechanismwithwhichscanverifythatitssourceroutewasused.Foraccountability,eachnodemightberequiredtosignthepacketsitforwards,butthisstrawmanap-proachveri esonlythatthepacketvisitedatleastthosenodesthatsignedit.Inthissection,thecentralchallengeistoallowasourcetocompelthenodesonitschosenpathtoforwarditspacketsalongonlythatchosenpath,evenwhenmorepro tablealternatepathsexist.2.2VCGRoutingMechanismWe rstreviewtheVickrey-Clarke-Groves(VCG)mech-anism[6,9,19].Eachtransitnodeiknowsthecostc(i;j)tosendtonodejwhere(i;j)2E.Asourcesroutestoadestinationdby rstrequestingfromeachtransitnodeitscosts,c(i;?).Inresponse,iclaimscostc0(i;j)forall(i;j)2E.ThegoalofVCGistoensurethateveryc(i;j)=c0(i;j):thateachireportsitscoststruthfully.LetSCP(s;d;H)denotetheshortestcostpathfromstodingraphH,C(SCP(s;d;H))thesumoftheedgecostsonthepath,andletHidenotethegraphinducedbyre-movingifromH.VCGaccomplishestruthfulreportingofcostsbyemployingthefollowingpayoutfunctionfromstoi:ifiisnotontheshortestcostpath,andthusdoesnowork,ps(i)=0,otherwise:ps(i)=C(SCP(s;d;Gi))C(SCP(s;d;G))+c0(i;j)(1)WesummarizeVCGwiththepseudocodeinAlgorithm1.Toseewhythisinducestruthfulreportingonbehalfofeachtransitnodei,observethatishouldneitherunder-statenoroverstateitscost:ishouldnotunderstate:Byde nition,C(SCP(s;d;Gi))C(SCP(s;d;G)),soeachnodeontheshortestcostpathreceivesatleastitsstatedcosttoforward.Ifc0(i;j)1 c(i;j),theniisnotguaranteedtorecoveritsactualcosttoforward,henceishouldensurethatc0(i;j)c(i;j).ishouldnotoverstate:ps(i)isindependentofc0(i;j)becauseC(SCP(s;d;G)),inEq.(1),includesc0(i;j).In-creasingc0(i;j)�c(i;j)could,however,removeifromtheshortestpath,soishouldensurethatc0(i;j)c(i;j).Algorithm1StandardVCG1.Gatherc0(i;j)fromeachiforeach(i;j)2E.2.ComputetheshortestcostpathPfromstodusingDijkstra'salgorithm.3.ForeachintermediatehopionP,payps(i)fromEq.(1)toi.2.3VCGwithBoycottingMechanismCulled­VCG.SupposesourceswishestoboycottasetofnodesBs.OnecouldrunVCGonthegraphinducedbyremovingallboycottednodes,GBs;werefertothisasculled-VCG.ThisapproachissimilartoMIRO[20],whichemployssource-routingtoallowfornodestovisit(oravoid)nodesoftheirchoosing.WeshowinFigure1,however,thatcullingtheboycottednodealonedoesnotsuce.Init,nodesissendingtonodedbutwishestoboycottB.TheshortestcostpathcomputedbystandardVCGisfs;A;B;C;dg,andbyculled-VCG,fs;A;C;dg.A,however,couldgainmorepro tbytunnelings'spack-etsthroughBatacostof2+,givingBpro t,andgivingApro t(1)inadditiontothepro tmadefroms'spayment.Culled-VCGfailsbecause,althoughnodeshaveincentivetotruthfullyreporttheircosts(thankstoVCG),itprovidesnoincentivefornodestoforwardonthepaththesourcespeci ed.VCGwithBoycotting.Wecandrawfromculled-VCGthefollowingobservation:foreachedge(i;j)intheshortestcostpathcomputedbyculled-VCG,ifthereisalower-costpathfromitojthatincludesaboycottednode(suchasthe(A;C)edgeinFigure1),thenthepacketswillmostlikelygothroughtheboycottednode.Weintroduceamodi edVCGmechanisminAlgorithm2.2.4TheCostofBoycottingBoycottinganodeislikelytoincuradditionalcost.WhensappliesAlgorithm2toFigure1,theresultingpathisfs;X;Y;dg,withpayoutsps(X)=ps(Y)=4,andc(s;X)=2,foratotalcosttosof10.WerestoapplystandardVCG,thepathwouldhavebeenfs;A;B;C;dg,withpayoutsps(A)=ps(C)=3,ps(B)=2,andc(s;A)=1,foratotalcostof9.Wede nethecostofboycottingtobetheratioofthetotalpricepaidbyswhenrunningVCGwithboycotting(Alg.2)tothepricepaidwhenrun-ningstandardVCG(Alg.1).IntheexampleofFigure1,shasacostofboycottingof10=9whensendingtodandboycottingB.Ifsvaluesboycottingin nitely,thensiswillingtoac-ceptanycostofboycotting,evenatthecostofnotbeingAlgorithm2VCGwithboycotting1.Gatherc0(i;j)fromeachiforeach(i;j)2E.2.Computeall-pairsshortest-costpathsontheweightedgraphG=(V;E;C),withtheweightofedge(i;j)equaltoc0(i;j).3.LetG0=(V;E0;C),whereE0=f(i;j)2Ejtheleast-costpathfromitojdoesnotcontainanyb2Bsg.4.Computetheleast-costpathPfromstodusingDijkstra'salgorithmonG0s(i.e.,runculled-VCGonG0).5.ForeachintermediatehopionP,payps(i)fromEq.(1)toi.abletoreachthedestination.However,if,foreachb2Bs,sassignssome niteutilityu(b)toboycottingb,thensmaynotbewillingtoboycottifthedi erenceinpriceisgreaterthanu(b).Insuchasituation,smaybewill-ingtoacceptsomesubsetSofBstoboycottsuchthattheadditionalcostincurredtoboycottSislessthantheutilitysgains,Pb2Su(b).Weleavecombinatorialboy-cottingpolicies(e.g.,\sendtononodesinSBs,ornomorethan3nodesfromTBs")tofuturework.3.BEHAVIORINANETWORKWITHBOY­COTTINGBeyondtheroutedeterminationgame,therearead-ditionalstrategiesthatbothboycottersandboycottednodesmustmake.Aslongasboycottersdonotvalueboycottingin nitely,boycottednodesmaymakeratio-nal,competitiveresponses.First,atransitnodeicanattempttodetermineifitisbeingboycotted,why,andtheimpacttoi'spro ts.Second,imayattempttoregainmarketshare,eitherviapolicychange(soastostopbeingboycotted)orbyextortingthosewhoarenotboycotted.Weaddresstheseinturn,and nishthissectionwithadiscussionofhownodesareexpectedtoreporttheircosts.3.1KnowingYourBoycottersRoutingentityimaywishtolearnaboutusers'boycottsets.Forinstance,imayask:(1)Areusersboycottingmeduetoaspeci cpolicyIdo(ordonot)exhibit?(2)Howmanyusersareboycottingme,andhowmuchpro tamIlosingasaresult?Boycottednodesmay nditextremelydiculttoan-swerthesequestionswithoutbeingabletoviewthetraf- cthattheywouldotherwisecarry.Inawirednetwork,wheniisboycotted,itreceivesnopackets,andhence,withoutaprioriknowledgeoftracdistributions,hasnowayofinferringhowmuchtracitisnotseeing.Inomnidirectionalwirelessnetworks,suchas802.11,eachtransmissionise ectivelyalsoabroadcast,soicouldper-hapslearnmoreoftheexistingtrac\rows,butonlyinclosephysicalproximity.Still,ihasseveraltoolsatitsdisposal:theobservedamountoftracandpayments2 411113422211311442223442221112 +dXZAACBsdYXZCBsdYXZ(a)(b)(c)(d)ACBsdYXZACsdYFigure1:(a)Examplenetwork:swishestoboycottB.(b)EdgesthatwouldinvolveBareculled.(c)Theshortestpathintheresultinggraphischosenwithcostofboycotting10=9.(d)AnexamplewhereBcouldextortA.ithasreceivedwithagivenc0(i;?),andthestated(andhencetruthful)costsofothers'edges.Ourinsightisthaticanperiodicallyperform\marketresearch"byloweringitsstatedcosts1andcomparethetracitisaskedtoforwardduringthelower-costpe-riodtonormalperiods.Suppose,forinstance,ireceivespacketswithsource-destinationpair(s;d)onlywhenitstatesalowercost,butiisontheleast-costpathfromstodevenwheniistruthful.Thenihasrevealedaninstancewheresisboycottingibut,uponstatinglowercost,iinducesacostofboycottingthatistoohighforstocover,atwhichpointitisnoteconomicallyfeasibleforstoboycotti.3.2ElicitingChangeWhywouldaboycottednodehavetoperformthismar-ketresearch?Wouldaboycotternotwanttobeoutspo-kenastowhomandwhyitisboycotting?Indeed,boycottingisamethodtoelicitpolicychange.Ifenoughend-usersboycottatransitnode,themessageshouldbeclear:changepolicytoearngreaterpro ts.Boycottingis,inasense,ameansofresolvingthetus-slebetweenend-userswhowantnodesthatforwardtheirpacketstoenforceaparticularpolicy(orlackthereof)andtheroutingentitiesthatdoordonotsupportthatpol-icy.Todoso,boycottersmustmakecleartotheroutingentitieswhytheyarebeingboycotted.Intwocases,however,anodesmaywishto\silently"boycottsomeroutingentityi.First,smaybesendingcon dentialmaterial,usingtheboycottasameanstoavoidsnoopingortoaccessdiversepaths,anddoesnotwanttopubliclyannouncethatitisboycottingitodoso.Second,smaywishtohelptheboycottgathermo-mentumandbecomesigni cantbeforestatinggrievances,hopingtodenyitheabilitytoquelltheboycottwithhalf-measures.Hence,marketresearchaswehavedescribeditmayservepowerfulroutingentitieswhowishtolearnaboutgrowing,grass-rootsboycottinge orts.3.3ExtortingNon­BoycottedNodesAsanalternativetochangingitspolicy,routingentityimayextortother,non-boycottedroutingentitiestogainpro t.ConsidertheexampleinFig.1(d);nodeB'struecosttosendtoCis2+forsomepositive.Inthiscase,ourmechanism(Alg.2)wouldnotremoveedge(A;C),1perhapsforonlyashortperiodoftime,soasnottoincurtoogreatalossandthepathfs;A;C;dgwouldbechosen,withpaymenttoA,ps(A)=(65+3)=4.Awouldreceiveapro tof1(afterpayingitscostof3toforward).Bmaythreatentounderstateitscostto1(resultinginFig.1)unlessAforwardssomeamount1ofitspro ttoB.AhasincentivetopayB(andearn1)ratherthanallowBtounderstateitscost(andearn0).Ofcourse,byunderstatingitscost,Brunstheriskofnotcoveringitscosttoforwardtracforother(s;d)pairsforwhichBisontheleast-costpath.Ifthetotalpro tfromextortingisgreaterthanthelossinpro tfromunderstating,thenBhasincentivetoextort.Sourcenodesmaywishtopreventb2Bsfromextort-ingforthesamereasonsthatitwishestoboycottb:toprotectagainstbearningmoneyortomakeiteconom-icallyinfeasibleforbtoinspects'spackets.Topreventagainstextortionattacks,wemodifyAlgorithm2asfol-lows:Algorithm3VCGwithboycottingandanti-extortion1.Gatherc0(i;j)fromeachiforeach(i;j)2E.Ifi2Bs,ignorethestatedcostandsetitto0instead.2.Runsteps2{5ofAlgorithm2asstated.Theintuitionbehindthismodi cationisthatsoughttoprovisionfortheworst-casescenario,inwhichbthreatenstounderstateitscostsasmuchaspossible(tozero).If,insodoing,bwouldnotmodifythechosenpath,thenbcannotextortanynodesonthepath.3.4CostRevelationWeinvestigatewhetherAlg.3givesincentivetonodeitotruthfullystateitsedgecosts.3.4.1BoycottedNodesSupposei2Bs.Asdiscussedabove,icouldemploypolicychangetotrytoremoveitselffromBs,butcanigainanyutilityfrommodifyingitsstatedcost?Evenifiweretoknowthatheisbeingboycottedbys,theanswertothisquestionis\no"byconstruction;the rststepstakesistoignorei'sstatedcosts.Thisremoveseventhepossibilityofextortingothernodes,asihasnoleverageovers'sroutingdecisions.Thus,withrespecttoanynodeswithBs3i,i'sstrategysetise ectivelynil,andtruth-tellingisaweaklydominantstrategy(therearenootherstrategiesthatimprovei'spayout).3 3.4.2Non­BoycottednodesOurmechanismispowerfulinthatitisimpervioustoanycountermeasuresmadebytheboycottednodes.Weshowherethat,surprisingly,itisthenodesthatarenotboycottedwhomayhaveincentivetolieabouttheircosts.Fixnowani62Bs,andletusconsiderunderstatingandoverstatingc(i;j)separately.UnderstatingCosts.AsinstandardVCG,foranyi62Bs,ishouldnotunderstateitscosts.Considertheoutcomesofsettingc0(i;j)c(i;j).SinceourmechanismusesthesamepayoutsasVCG,wehavethesamepropertyoutlinedinSection2.2;icannotcontrolthepaymentthatitreceives:onlywhetherornotitisontheshortestcostpathfromstod,SCP(s;d;G).Ifbystatingitstruecostiwouldhavebeenontheshortestpathfromstod,thenigainsnoadditionalutilityfromunderstatingitscost.IfbystatingitstruecostiwasnotonSCP(s;d;G),thenstatingac0(i;j)c(i;j)couldputiontheshortestpath.However,asdescribedinSection2.2,theonlyassuranceihasisthatps(i)c0(i;j),henceihasnoguaranteethatitscostofforwardingwouldberecovered,andmightthereforedecreaseitsutility.OverstatingCosts.Suppose rstthatbytruthfullyre-portingitscost,iwouldbeonSCP(s;d;G).Thenbyoverstatingitscost,iwouldrisknotbeingontheshortest-costpath;further,asabove,evenifitremainedonthepath,itwouldnotincreaseitspro t.InthecasewheretruthfulreportingwouldnotplaceionSCP(s;d;G),theproofofstandardVCGgoesasfollows:increasingitscostcannotmakeitspathshorter,andhenceicangainnoutilityinoverstating.However,inourmechanism,therearetwocasesinwhichidoesnotappearinSCP(s;d;G):Inthe rstcase,noneofi'sedgeswereculled;i'spathsimplycoststoomuchtobeconsideredbys,ashappenedwithnodeZinFig.1.AsinstandardVCG,suchanodeexperiencesnochangeinutilityfromoverstating.Inthesecondcasewhere,fori62SCP(s;d;g),iwouldhavebeenontheshortestpathbutatleastoneofitsedgeswasculledtoavoidaboycottednode,aswithnodeAinthe gure.IfAweretohaveoverstateditscosttoB2Bstosomevaluegreaterthan2,thenswouldnothaveculledthe(A;C)edge.Aswiththeextortionattack,nodeAwillonlyhaveincentivetooverstateitscosttoBifthepro tgainedfromnodesboycottingBisgreaterthanthepro tlostfromthosewhowouldhavegonethroughAtoBiftheedgecostless.Themostlikelyscenarioinwhichthisoccursiswhenamajorityofend-usersboycottB.3.4.3ASybilAttackAtransitnodecanincreasetheincentivethatithastooverstateitscoststoboycottednodeswiththefollowingSybilattack[7].SupposenodeihasnneighborsingraphG.Toensurethatiisaviabletransitnodeforanyend-usersforwhichiandatleasttwoofi'sneighborsarenotinBs,icancreaten2pseudo-identities(\Sybils"),oneforeachpairofneighbors.EachsuchSybilconnectsonlythatpairofneighbors;weshowanexampleinFigure2.11313111311AsBAssdBsddBAABdFigure2:Acreates3Sybils,atleastoneofwhichhasanedgetodthatswillnotcullwhileboy-cottingB.VariousschemesprotectagainstSybilattacks,suchaschargingeachparticipantanentryfee.Wedonotattempttoprovideasolutionhere,onlytopointoutthattransitnodesmayhaveincentivetoperformaSybilattackandacounter-measuremaybenecessarytosupportboycotting.AnISPisunlikelytobeabletolaunchsuchaSybilattack.3.5AddressingtheOverstatingofCostsUnfortunately,ourmechanismdoesnotensurethatnon-boycottednodeswilltruthfullyreporttheiredgecosts.Althoughthisholdsonlyinextremecases|whentheboy-cottednodeisboycottedbysomanyend-usersthatin-creasingitsstatedcostwouldnotcausealossofpro t|wewouldliketohaveamechanismthatalwaysgivesincentivefortruthfulness.Webelievethatthereisnopaymentschemethatwillprovidethisincentive,butareunabletoprovideaproofatthistime.Addressingover-statingisthemostimportantopenprobleminthislineofwork.Weareconsideringheuristicstoaddressthisproblem,suchas:(1)IfatanypointAisfoundtobeoverstatingitscosts,orifitisfoundtobeforwardings'spacketsthroughsomenodeb2Bs,thensboycottsA,aswell.(2)scouldrequirethatAproveitscoststob.4.SIMULATIONRESULTSToobservethecostofboycotting,weevaluatedAlgo-rithms2and3withasimulator.Ourpreliminaryresultsindicatethatifanodecanbeboycotted,itcanbeboy-cottedcheaply.Weevaluateourmechanismbothinadhoctopologies(totestitongeneral,shortest-costpathnetworks)andonasubsetoftheInternetAStopology(whichallowsonlypolicy-compliantpaths).WirelessAdHocTopology.Thesimulationsuseda xedtopologyof130nodesthatwastakenfromawirelessadhocnetworkinPortland,OR[3].Nodeswithin250me-tersofeachothersharedanedge.Weassignededgescostsbetween1and20fromaZipfdistribution(mostedgeshavesmallercost).WechosetheZipfdistributiontomodelanetworkwithmanyhomogeneouslinksthatallowmostnodestocommunicatecheaplyandasmallnumberofmoreimportantlinkswhoseservicecomesatapremium.Foreachhsource,destinationipair,chosenatrandom,wecalculatedthecostoftheshortestpathfromthesourcetothedestinationusingVCG.Then,wesimu-latedthesourceboycottingeachnodealongthatshortestpathindividually,andwerantheboycottingmechanism4 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0.1 1 10Cumulative fractionRatio to standard VCG paymentBoycottAnti-extortion 0 0.2 0.4 0.6 0.8 1 0 0.5 1 1.5 2 2.5 3 3.5 4Cumulative fractionRatio to shortest path costBoycottAnti-extortionFigure3:Comparingtheboycotting(Alg.2)andboycottingwithanti-extortion(Alg.3)mechanismstostandardVCG.Left:sumpaymentsmadebythesourcenode,normalizedtostandardVCG.Right:costofshortest-costpath,normalizedtostandardVCG.ThecostdistributionisZipf.andtheboycottingwithanti-extortionmechanism.WepresentourresultsinFigure3.VCGrequiresatleasttwopathsbetweenagivensourceanddestination(inordertocomputethemarginalbene tofeachedgeontheshortest-costpath),andourproposedmechanismsgenerallyrequiremore(toallowforcullededges).Otherwise,themechanismwillreturnanin nitecost.InthesamplerundepictedinFigure3,VCGre-turneda nitecostfor62.7%ofthehsource,destinationipairs,VCGwithboycottingreturneda nitecostfor51.6%ofthehsource,destination,boycottednodeitu-ples(10.6%ofthegraphsweredisconnected),andVCGwithboycottingandanti-extortionreturneda nitecostfor38.6%oftuples(17.2%ofgraphsweredisconnected).TheplotsinFigure3aretakenacross1082hsource,destination,boycottednodeituples,andarenormalizedtothestandardVCGcosts(whenthesourceisnotboy-cotting).IntermsofbothVCGcostandpathlength,thecostofboycottingandthecostofboycottingwithanti-extortionaresimilar.Webelievethisisbecausepathsarelikelytoincludecheapedges.Whenanodeonthepathisboycotted,itislikelythatthecostofitsoutgoingedgesisnearly1,sosettingthemto0(Alg.3)haslittlee ectonwhetherornotthatpathisculled.Forabout25%ofpaths,thepathcostischeaperwhenanodeisboycotted.Thisispossibleinthefollowingscenario:thereisasinglecheappathfromthesourcetothedestination,andseveralmoreexpensivepathsthathavesimilarcoststoeachother.Whenanodeonthecheappathisboycotted,amoreexpensivepathmustbechosen.Sincetheexpensivepathsarepricedsimilarly,themarginalbene tofoneovertheotherislessthanthemarginalbene tofthecheappathovertheexpensivepaths.Inthiscase,VCGmayreportthatthecostofboycottingischeaperthanwhennoboycottingisdone.Thismayresultinsendersboycottingcertainnodestodecreasethecostoftheirpaths.InternetASTopology.Wealsoranexperimentstotrytodeterminehowfeasibleitwastoboycott\majorplayers"intheInternet.Forthissimulation,weusedagraphofthe1170nodeswithhighestdegreeintheInternetAStopology[18].Wesettheedgecostsofnodeitoblog2(degree(i))c+1,tomodelhigher-degreenodesasbe-ingmoreexpensive,buttakingthelogarithmtolimitthecostdisparity.Tomakesureourresultsdidnotdependonthechoiceofcostfunction,wealsoranseparateex-perimentssettingtheedgecostsofnodeitodegree(i)andtob1024=degree(i)c.Boycottingmany(upto15)tier-1ISPssimultaneously,nodeswerestillabletomain-tainconnectivityforalmostallrandomlychosenhsource,destinationipairs,forallmodelsofcostfunctions.How-ever,ourresultalmostcertainlyoverestimatesthecon-nectivitysinceweassumethatallremaininglinkscanbeusedforforwardinganypacket(regardlessoforigin).Inreality,manylinkswillbesubjecttoAS-localpolicy,andmaynotbeusedtoforwardthird-partytrac.5.IMPLICATIONSINTHEINTERNETTheboycottingmechanismwedescribedisnotfeasibleintoday'snetwork.Althoughtherearemechanismsforchoosingpaths(loosesourcerouting)andfordeterminingwhichAS-pathanext-hopISPmaychoose,boycottingisinfeasibleforthreereasons.First,sourcesdonot(andhavenomechanismto)payentitiesbeyondtheirneigh-bors.Paymentisinsteadachievedthroughlongstanding,pairwiseagreementsthatareunlikelytochangewhenaboycottbegins.Second,thecostofeachlinkiskeptse-cret,andmayvary,whichmeanssourcescannotreliablyestimatethe\cost"ofpreferringonepathoveranother,anestimatethatwouldbeneededtocompensatethenet-workimplementingtheroutechoice.Third,thepathapackettakescannotbeveri ed:aprovidermightacceptpaymentforboycottinganupstreamAS,butproceedtouseitanyway.Thee ectivenessofaboycottderivesfromaquanti -ablelossinpro tthatcanbetracedtoaspeci ceventorgroupofaggrievedsources.Eithersuddenlyshiftingtracawayfromaboycottednodethreatenscongestion,oritrisksbeingnulli edbytracengineering.Overlays,andperhapsloosesourcerouting,provideamechanismforuserstobypassnetworkroutingtoemulateboycottpolicies.Toboycottmanynodeswouldrequiremanycooperatingoverlayparticipants.Toboycottwithhighvolumesoftracmaybeinfeasible,forperformance(forwardingmaybedicult)andbecauseISPtracen-5 gineersmayreadjusttracawayfromnewly-congestedlinks.Wewonderwhetheranoverlayroutingapproachcan(andshould)adaptquicklyenoughtocounteranytracengineeringinthemiddleofthenetwork.WithinBGP,oneASmaybeabletopartiallyboycottanotherbyinspectingpathsadvertisedbydi erentup-streamprovidersandpreferringtheproviderwhoseroutesdonotincludetheboycottedISP.ThepartialboycottdoesnotchangeBGPinanyway.MIRO[20]wouldpro-videadditionalpowertothisapproachbyallowingastubAStochooseamongpathso eredbyitsprovider.The\cost"ofthedecisionislocal,inthattheASchangingroutestoe ectaboycottisnolongeroptimizingthepriorpathchoice.BeyondBGP,inter-domainroutingproposalshavebeenproposedtoallowusersortheirimmediateproviderISPstoexercisegreatercontroloverthepathstraversedbytheirpackets.Tracebackapproacheswouldallowuserstoverifythepathtakenbytheirpackets.Thecombina-tionofsourcecontrolandveri cationproposedmaybesucientforimplementinganyboycottingschemeinafutureInternet.Thepotential\rexibilityo eredbysuchnetworksmotivatesourstudyofabstractVCGnetworks.6.RELATEDWORKResearchershaveappliedVCG-like[6,9,19]mecha-nismstoimproveroutingindistributedenvironments.NisanandRonen[14]provedthefeasibilityofstrate-gyproof,polynomialtimeVCGmechanismsfornetworksofsel shusers.Thisresultmotivatedsigni cantwork[1,8,11,15].AndereggandEidenbenzproposeAdhoc-VCG[1],aprotocolto ndminimum-energypathsinmo-bileadhocnetworkswithsel shparticipantsthatacceptpaymentsforforwardingdata.Feigenbaumetal.[8]builduponBGPtocomputeleast-costASpaths.TheyshowthatadistributedimplementationofVCGdoesnotre-quireconsiderablecost.Thesealgorithmscomputepathsthatareshortestbyasinglecostmetricanddonotcon-sideruser-speci edcriteriaforchoosingpaths.Providinguserswithautonomytochoosepathshasbeenproposedasameanstocircumventfailures[2]andimproveperformance[10,17].Usersoradministratorsofsmallernetworksgainthisautonomythroughsourcerout-ing[16,21],overlays[2,17],oralterationstoBGP[13,20].Thesemechanismsmaybeusefultowardimplement-ingaboycott,buttheresearchhasbeenfocusedonmorequanti ablebene tsinperformanceandreliability.Inwirelessnetworks,blacklistingcanpreventlowqual-itylinksornodesfrombeingconsideredinthepathselec-tionprocess[4,5,12].However,blacklistingisnotdoneattherequestofthesender:anodeorlinkisremovedfromthenetworkifitsperformance,bysomemeasure,fallsbelowaprede nedthreshold.7.CONCLUSIONSWeintroducedtheboycottasapolicygoalapplicabletoanetwork.WeshowedhowtoadapttheVCGroutingmodeltosupportboycottinggroupsofnodesandfoundthattruthfulreportingoflinkcostsremainsaweaklydominantstrategywhenboycottingisinfrequent.WesimulatedtheroutingpoliciesonadhocnetworkandAS-graphtopologiestoshowthatthecostofboycottingistypicallysmallandthatdisconnectednessisrare.Manyopenproblemsandresearchdirectionsremain.Currently,nodeshaveincentivetooverstatetheircoststoboycottednodes(e ectivelydisassociatingthemselves),albeitonlywhentheboycottingiswide-spread.Ourworkassumesshortest-costrouting,butBGPrequirespathsto(also)bepolicy-compliant.Extendingourmechanismtoworkinthissettingiscrucialtounderstandingitsfeasi-bilityintheInternet.Also,webelieveboycottingmaybeusedasameansofsimulatingandestimatingthenet-work'sresiliencetocatastrophicfailure.8.ACKNOWLEDGMENTSThisworkwaspartiallysupportedbyNSFAwardsCNS-626636,CAREERANI-0092806,andafellowshipfromtheSloanFoundation.DaveLevinwasalsosupportedinpartbyaUMDCSDDeansfellowship.9.REFERENCES[1]L.AndereggandS.Eidenbenz.Adhoc-VCG:Atruthfulandcost-ecientroutingprotocolformobileadhocnetworkswithsel shagents.InProc.ofMobiCom,2003.[2]D.G.Andersen,H.Balakrishnan,M.F.Kaashoek,andR.Morris.Resilientoverlaynetworks.InACMSOSP,2001.[3]H.Balakrishnan,etal.TheDistance-2MatchingProblemanditsRelationshiptotheMAC-LayerCapacityofAdHocWirelessNetworks.IEEEJSAC,22(6):1069{1079,2004.[4]M.Busse,T.Haenselmann,andW.E elsberg.Energy-ecientforwardingschemesforwirelesssensornetworks.InWoWMoM,2006.[5]K.-W.Chin,J.Judge,A.Williams,andR.Kermode.ImplementationexperiencewithMANETroutingprotocols.SIGCOMMCCR,32(5):49{59,2002.[6]E.Clarke.Multipartpricingofpublicgoods.Publicchoice,11,1971.[7]J.Douceur.TheSybilAttack.InIPTPS,2002.[8]J.Feigenbaum,C.Papadimitriou,R.Sami,andS.Shenker.ABGP-basedmechanismforlowest-costrouting.InPODC,2002.[9]T.Groves.Incentivesinteams.Econometrica,41,1973.[10]K.Gummadi,etal.ImprovingthereliabilityofInternetpathswithone-hopsourcerouting.InOSDI,2004.[11]J.HershbergerandS.Suri.Vickreypricesandshortestpaths:Whatisanedgeworth?InFOCS,2001.[12]H.Lundgren,E.Nordstro,andC.Tschudin.CopingwithcommunicationgrayzonesinIEEE802.11bbasedad-hocnetworks.InWoWMoM,2002.[13]R.Mahajan,D.Wetherall,andT.Anderson.Negotiation-basedroutingbetweenneighboringISPs.InNSDI,2005.[14]N.NisanandA.Ronen.Algorithmicmechanismdesign.InACMSymposiumonTheoryofComputing,1999.[15]D.C.ParkesandJ.Shneidman.DistributedimplementationsofVickrey-Clarke-Grovesmechanisms.InACMAAMAS,2004.[16]B.RaghavanandA.C.Snoeren.Asystemforauthenticatedpolicycompliantrouting.InSIGCOMM,2004.[17]S.Savage,etal.Detour:InformedInternetroutingandtransport.IEEEMicro,19(1):50{59,1999.[18]http://www.caida.org/data/active/as-relationships/.[19]W.Vickrey.Counterspeculation,auctions,andcompetitivesealedtenders.JournalofFinance,16,1961.[20]W.XuandJ.Rexford.MIRO:Multi-pathinterdomainrouting.InACMSIGCOMM,2006.[21]X.Yang.NIRA:AnewInternetroutingarchitecture.InFDNA,2003.6