Abstract---Peer-to-peer computing and networking, a new model of communication and computation, has recently started to gain significant acceptance. This model not only enables clients to take a more active role in the information dissemination process, bu
Tracingalarge-scalePeertoPeerSystem:
anhourinthelifeofGnutella
EvangelosP.Markatos
InstituteofComputerScience(ICS)
FoundationforResearch&Technology–Hellas(FORTH)P.O.Box1385,Heraklio,Crete,GR-711-10GREECEdoc.guandang.net
TechnicalReport298
ICS-FORTH
appearsinCCGrid2002:thesecondIEEEInternationalSymposiumonClusterComputingandtheGrid
Abstract—Peer-to-peercomputingandnetworking,anewmodelofcom-municationandcomputation,hasrecentlystartedtogainsigni cantaccep-tance.Thismodelnotonlyenablesclientstotakeamoreactiveroleintheinformationdisseminationprocess,butalsomaysigni cantlyincreasetheperformanceandreliabilityoftheoverallsystem,byeliminatingthetradi-tionalnotionofthe“server”whichcouldbeasinglepointoffailure,andapotentialbottleneck.
Althoughpeer-to-peersystemsenjoyasigni cantandcontinuallyin-creasingpopularity,westilldonothaveaclearunderstandingofthemagni-tude,thetraf cpatterns,andthepotentialperformancebottlenecksoftherecentpeer-to-peernetworks.
Inthispaperwestudythetraf cpatternsofGnutella,apopularlarge-scalepeer-to-peersystem,andshowthattraf cpatternsareveryburstyevenoverseveraltimescales.WeespeciallyfocusonthetypesofthequeriessubmittedbyGnutellapeers,andtheirassociatedreplies.Weshowthatthequeriessubmittedexhibitsigni cantamountsoflocality,thatis,queriesarefrequentlyandrepeatedlysubmitted.WeproposesimpleGnutellacachingmechanismsthatcachequeryresponses.Usingtrace-drivensimulationweevaluatetheeffectivenessofGnutellacachingandshowthatitimprovesperformancebyasmuchasafactoroftwo.
I.INTRODUCTION
Recently,anewmodelofcommunicationandcomputation,calledpeer-to-peernetworking,hasstartedtogainsigni cantacceptance[1].Contrarytothetraditionalclient-servermodel,peer-to-peercomputingenablesallclientstoactasserversandallserverstoactasclients.Inthisway,clientsnotonlytakeamoreactiveroleintheinformationdisseminationprocess,butalsomaysigni cantlyincreasetheperformanceandreliabilityoftheoverallsystem,byeliminatingthetraditionalnotionofthe“server”whichcouldbeasinglepointoffailure,andabottleneckintheoverallsystem.
The rstandmostwidely-knownpeer-to-peersystem,Nap-ster,1isa lesharingutilitythathasenabledhundredsofthou-sandsofuserstoef cientlyshare les,includingmp3-encodedsongs,overtheInternet.CapitalizingonthesuccessofNap-ster,severalotherpeer-to-peer lesharingsystemshavebeenrecentlydeveloped.TheseincludeGnutella,KaZaA,Audio-Galaxy,etc.Althoughthetechnicaldetailsofthesesystemsvarysigni cantly,theyallsharethepeer-to-peerphilosophybyenablingallpeerstostoreanddelivercontenttootherpeers.Be-sides lesharing,peer-to-peersystemshavealsobeenusedfor
doc.guandang.net cientexecutionofhighlyparallelanddistributedapplica-tionsbycapitalizingontheavailabilityofidlecyclesinhomecomputers.Suchapplicationsrangefromsystemsthatsearchforextra-terrestrialintelligence[2],tosystemsthatsearchacureforAIDS[3].
Althoughpeer-to-peersystemsappearedrelativelyrecently,theirpopularityhasincreasedrapidlyinthelastcoupleofyears.Forexample,networktraf cmeasurementsattheUniversityofWisconsinsuggestthatintheperiodofApril2000,Napster-relatedtraf crepresentedthe23%oftheirtotalnetworktraf c,whileatthesametimeweb-relatedtraf caccountedforonly20%[4],[5].AlthoughNapstertraf chasbeenreducedrecently,thepercentageofallpeer-to-peertraf chasactuallyincreased.Forexample,recentmeasurementfromtheUniversityofWisconsinsuggestthat,inOctober2001,peer-to-peertraf creachedmorethan30%ofthetotaltraf c2,whileatthesametime,web-relatedtraf cwasalittlemorethan19%.
Althoughthesemeasurementssuggeststhatthetraf cde-mandsofpeer-to-peersystemsrepresentasigni cantandcon-tinuallyincreasingpercentageoftheoverallnetworktraf c,westilldonothaveaclearunderstandingofthemagnitude,thetraf- cpatterns,andthepotentialperformancebottlenecksofsuchpeer-to-peernetworks.
Inthispaperwestudythetraf cpatternsofGnutella,apopularlarge-scalepeer-to-peersystem.WeespeciallyfocusinthetypesofthequeriessubmittedbyGnutellapeers,andtheirassociatedreplies.WeidentifythelocalitypatternsthatexistinGnutellaqueriesandproposesimple,buteffectivecachingmechanismstoexploitthem.
Moreaccurately,thecontributionsofthispaperare:
Wemodi edanopen-sourceGnutellaclient(gnut[6])andproduceddetailedtracesofthequeriesreceived(andforwarded)byourclient.
Weinstalledthreesuchclients,intwodifferentcontinentsandconcurrentlygatheredtracesofGnutellaqueries.
Westudiedthetraf cduetoGnutellaqueryrequestsandqueryresponsesandfoundthatthetraf cisburstyandcontinuestoremainburstyoverseveraltimescales.
WestudiedtheGnutellaquerytracesandshowedthatthey
doc.guandang.net
Iamnitchi. Mapping the Gnutella Network: Properties of Large-Scale Peer-to-Peer Systems and Implications for System Design. IEEE Internet Computing Journal,...
The system is based on the tracing of the spending pattern of coins, and...[Coh03] and Gnutella [Kir] are distributed systems without centralised ...
Tracing a large-scale pe... 暂无评价 9页 ...The most successful decentralized system is Gnutella...server model than a true peer-to-peer system. ...
The system is based on tracing the spending ...[Coh03] and Gnutella [Kir], and grid systems ...If such large scale computations are for an ...
Tracing a large-scale Peer-to-Peer System: an hour in the lifetime of Gnutella. 2nd IEEE/ACM Int. Symp. On Cluster Computing and the Grid 2002 B...
The system is based on tracing the spending ...[Coh03] and Gnutella [Kir], and grid systems ...If such large scale computations are for an ...
热门文档