















LiveandcollectingBGPASpathstowardsthespamnetworkfromsixRouteViewsBGPcollectorsdistributedworldwide.Additionally,wemapeveryIP-levelhopintraceroutestotheASannouncingitandfurtherenrichthepathwithgeolocationinformation[3]andtheregisteredholder[11]ofeachIP-andAS-levelhoptraversedbyatraceroute.
Tosummarize,everyspamIPaddressblockmonitoredover7consecutivedaysisassociatedwith:
asetofdailyIP/AStraceroutepathsfromourvantagepointtothenetwork;
asetofdailyBGPASpathsfromthesixRouteviewsBGPcollectorstothenetwork;
additionalinformationonthegeolocationandownerofeachintermediateIP-andAS-levelhopintracer-outes.
B.Multi-stagescoringanddata ltering
AstheamountofIPaddressblockstomonitorincreasedsigni cantlyovertime,weneededamechanismtoautomat-icallyinvestigatethem.Thisiswhywedesignedamulti-stagescoringand lteringsystemthatanalyzestherawdata,identi esabnormalroutingevents,assignsindividualscoresbasedonaconsistentsetofcriteria,andthenaggregatesallscorestoeventuallyhighlightIPblocksmostlikelyindicatingpossibleBGPhijacks.Wedescribeherethemaincomponentsofthismulti-stagescoringand lteringsystem.
1)TracerouteandBGPanomalyscoring:In[47]weintroducedasetofheuristicsthatprovedtobeeffectiveatidentifyingBGPandTracerouteanomalies.Withthelimiteddatawehadatourdisposalatthattime(April-Sep2011),thisapproachhadonlyunveiledalimitednumberofcasesinvolvingafewIPblocksbeinghijacked.Inthispaper,weapplyittoamuchlargerdatasetandrevealasigni cantlylargeramountofsuccessfulhijacks.
(I)BGPanomaliesprovideaviewfromthecontrolplaneontheroutingbehaviorofmonitorednetworksandareex-tractedfromthesetofdailyBGPASpaths.(I.a)ABGPoriginanomalyreferstoanIPaddressblockbeingannouncedbymorethanoneAS.SuchanomalyisalsocommonlyreferredtoasaMultipleOriginAS(MOAS)con ict.(I.b)ABGPASpathdeviationmeasuresthedifferenceobservedbetweenBGPASpathscollectedfromagivenBGPcollectortowardsagivenIPaddressblock.
(II)Tracerouteanomaliesareusedtoassesstheimpactofcontrolplaneroutingchangesonthedataplane.TheyareextractedfromthesetofdailyIP/AStraceroutepaths.(II.a)AnIP/ASreachabilityanomalyreferstoapermanentchangeinthereachabilityoftheprobeddestinationhostorAS.(II.b)AHopcountanomalyquanti esapossiblepermanentchangeinthelengthoftraceroutes.(II.c)AnIP-leveltraceroutedeviationandanAS-leveltraceroutedeviationmeasurethedifferenceobservedbetweenrespectivelyIP-leveltraceroutesandAS-leveltraceroutes.(II.d)AGeographicaldeviationquanti esthedifferenceobservedbetweenthecountriestraversedbytraceroutes.
Asdescribedin[47],everyanomalytypeisquanti edwithascorein[0,1].ABGPoriginanomalyisde nedbyaFig.1.Experimentalenvironment.
approachisthatwhenanIPaddressblockishijackedforstealthyspammingthenaroutingchangewillbeobservedwhentheblockisreleasedbythespammertoremainstealthy.Sincewestartmonitoringanetworkwhenwereceivespamfromit,welookforaroutingchangefromthehijackedstateofthenetworktothenormalstateofthenetwork.Thegoalhereisnottobuildastand-aloneBGPhijackdetectionsystembutinsteadtocollect,inreal-time,routingdataassociatedwithspamnetworksinordertoidentifyspammerssendingspamfromtemporarily(i.e.,lessthanoneday)hijackedIPaddressblocksasdescribedin[35].Intheremainderofthissectionwedescribethedifferentpartsofourexperimentalenvironmentinmoredetails.
A.Routingdatacollection
1)SpamIPaddressblockselection:Ourprimarydatasetisalivefeedofspamemailscollectedatspamtraps.Everydaywereceiveabout3,500,000spamemailsfromabout24,000distinctIPaddressblocks.DuetotheoverheadimposedbytraceroutemeasurementsandbyqueryingtheBGPcollectors,oursystemcancurrentlymonitorabout8,000IPaddressblocksonadailybasis.AsampleofIPaddressblocksisextractedfromthespamfeedeveryhour.Whenselectingblockstomonitorweprioritizetherecentlyannouncedonesastheyaregoodcandidatesforshort-livedhijacksassuggestedin[35].WeconsidertoberecentlyannouncedanyIPaddressblockinourspamdatasetthatbecameroutedwithinthelast24hours,basedonarchivedroutinginformationbases(RIBs)fromRouteViewsandRIPERIS.
2)TracerouteandBGPmonitoring:WemonitoreachspamIPaddressblockduringsevendaysbylaunchingtraceroutes
热门文档