Upload
others
View
2
Download
0
Embed Size (px)
Citation preview
DistributedSystemsandNetworkslabDepartmentofComputerScience,JohnsHopkinsUniversity
www.dsn.jhu.edu
YairAmirDonP.GiddensLecture
• Theoneandonly–MichalMiskin‐Amir• Theadvisor–DannyDolev• Theprofessors–
– M.Melliar‐Smith,L.Moser,K.Birman,A.Brodsky,Y.Yemini
• Thestudent‐colleagues‐– R.Borgstrom,J.Stanton,D.Shaw,J.Green,J.Schultz,T.Schlossnagle,A.Peterson– C.Nita‐Rotaru,C.Danilov,C.Tutu,R.Caudy,A.Munjal,M.Hilsdale– N.Rivera,J.Lane,R.Musaloiu‐EleNeri,J.Kirsch,M.Kaplan– D.Obenshain,T.TanQllo
• Thego‐toexperts‐– B.Awerbuch,A.Barak,G.Tsudik,S.Goose,A.Terzis,B.Coan,R.Ostrovsky
• Theentrepreneurs‐– M.Khan,Y.Javadi,S.Goose
• TheHopkinsprofessors‐– B.Awerbuch,G.Mason,R.Kosaraju,S.Smith,M.Goodrich,R.Westgate
• Theprogrammanagers–– D.Maughan,T.Gibson,C.Landwehr,H.Shrobe
2February16,2012 YairAmir
Asingle,mulQ‐purpose,IP‐basednetwork– EachaddiQonalnodeincreasesitsreachandusefulness(similartoanynetwork)
– EachaddiQonalapplicaQondomainincreasesitseconomicadvantage
– Willthereforeswallowmostothernetworks• Happened:mailtoe‐mail,PhonetoVoIP,FaxtoPDFs• Startedtheprocess:TV,variouscontrolsystems• SQlltocome:Cellphonenetworks
3February16,2012 YairAmir
Asingle,mulQ‐purpose,IP‐basednetwork• Theartofdesign–asuccessfulparadigm
– Keepitsimpleinthemiddle• Best‐effortpacketswitching,rouQng(intranet,Internet)
– Smartattheedge• End‐to‐endreliability,naming
• Couldthereforeadaptandscale– Survivedfor4decadesandcounQng– Sustainedatleast7ordersofmagnitudegrowth
• Standardizedandalotridesonit– Thebasicservicesarenotlikelytochange
4February16,2012 YairAmir
• CommunicaQonpaberns– FromPoint‐to‐point–topoint‐to‐mulQpoint–tomany‐to‐many
• Highperformancereliability– “Fasterthanreal‐Qme”filetransfers
• LowlatencyinteracQvity– 150mskeystrokemirroring– 100msforVoIP– 80‐100msforinteracQvegames(remotesurgery?)
• End‐to‐enddependability– From“Internet”dependability–to“phoneservice”dependability–to
“TVservice”dependability–to“remotesurgery”dependability• Systemresiliency
– FromE‐mailfaulttolerance–tofinancialtransacQonsecurity–tocriQcalinfrastructure(SCADA)intrusiontolerance
5YairAmirFebruary16,2012
• Buildspecializednetworks– WasdonedecadesbeforetheInternet– ThinkCable/TVdistribuQon(Satellite+lastmile)– Extremelyexpensive
• BuildprivateIPnetworks– AvoidstheresourcesharingaspectsoftheInternet,solvessomeofthe
scaleissues– Expensive– SQllconfinedtobasicIPnetworkcapabiliQes
• BuildabeberInternet– ImprovementsandenhancementstoIP(orTCP/IPstack)– “Cleanslatedesign”
• Buildoverlaynetworks
6YairAmirFebruary16,2012
• Overlay paradigm: – In contrast to “keep it simple in the middle and smart at the edge” – Move intelligence and resources to the middle
• Software-based overlay routers working on top of the internet • Overlay links translated to Internet paths
• Smaller overlay scale (# nodes) smarter algorithms, better performance, and new services.
February16,2012 YairAmir 7
• FlexibleRouQng– RON–resilientrouQngusingalternatepaths[Andersenetal,01]– XBone–flexiblerouQngusingIPinIPtunneling[Touch,Hotz,98]
• ContentDistribuQon– Yoid–host‐basedcontentdistribuQon[Francis00]– Overcast–reliablemulQcastforhighbandwidthcontentdistribuQon[Jano:
etal,00]– Bullet–mulQ‐pathdatadisseminaQon[Kos<cetal03]
• MulQcast– ESM–providesapplicaQon‐levelmulQcast[Chuetal,00]– HTMP–interconnectsislandsofIPMulQcast[Zhangetal,02]
• PeertoPeer– Chord–logarithmiclookupservice[Stoicaetal,01]– Kelips–O(1)lookupwithmoreinformaQonstored[Guptaetal,03]
• GroupCommunicaQon– TheSpreadtoolkit–scalablewideareagroupcommunicaQonusinganoverlay
approach[Amir,Danilov,Stanton,00]
February16,2012 YairAmir 8
• TheOverlayNetworkParadigm• TheDARPANetworkingChallenge(99‐03)
– OverlayArchitecture– Low‐latencyreliabletransport
• TheSiemensVoIPChallenge(03‐06)– Almost‐reliable,real‐Qmetransport
• TheLiveTimeNetTVChallenge(08‐…)– FromOverlaystoClouds– UlQmateresiliency,automatedmonitoringandcontrol
• ThechallengesaheadFebruary16,2012 YairAmir 9
• ThetradiQonalparadigm(keepitsimpleinthemiddleandsmartattheedge)workswellfortradiQonalapplicaQonsintypicalconnecQvitycondiQons
• ButnotsowellfortradiQonalapplicaQonsinbadconnecQvitycondiQons
• AndnotsowellforemergingapplicaQonsintypicalconnecQvitycondiQons
February16,2012 YairAmir 10
• 50millisecondnetwork– E.g.LosAngelestoBalQmore– 50millisecondstotellthesenderabouttheloss– 50millisecondstoresendthepacket
• Atleast100millisecondstorecoveralostpacket
February16,2012 YairAmir 11
565
5
LAX BWI
• 50millisecondnetwork– E.g.LosAngelestoBalQmore– 50millisecondstotellthesenderabouttheloss– 50millisecondstoresendthepacket
• Atleast100millisecondstorecoveralostpacket– Canwedobeber?
February16,2012 YairAmir 12
LAX BWI
• 50millisecondnetwork,fivehops– 10millisecondstotellnodeDALabouttheloss– 10millisecondstogetthepacketbackfromDAL
• Only20millisecondstorecoveralostpacket– LostpacketsenttwiceonlyonlinkDAL–ATL
February16,2012 YairAmir 13
56 5
5
LAX PHX DAL ATL DCA BWI
SimulaQon
Latency
February16,2012 14YairAmir
SimulaQon
Latency
Ji@er
February16,2012 15YairAmir
SimulaQon SpinesonEmulab
Latency
Ji@er
February16,2012 16YairAmir
• 50msnetworkdividedevenlyintoxhops• Delayedpackets:arriveaNermorethan50+10ms
February16,2012 YairAmir 17
• TheSpinesOverlayMessagingsystem– AnOverlaysoNwarerouter(daemon)ontopofUDP– RunningasanormalInternetapplicaQon
• Easytouseprogrammingplarorm– TransparentinterfaceidenQcaltothesocketinterface,givingTCP,UDPandIPMulQcastfuncQonality
• “Commercialgrade”deployablesystem– ImprovingapplicaQonperformanceovertheInternet– Enablingnewservices– Opensource(www.spines.org)
February16,2012 YairAmir 18
February16,2012 YairAmir 19
• TheOverlayNetworkParadigm• TheDARPANetworkingChallenge(99‐03)
– OverlayArchitecture– Low‐latencyreliabletransport
• TheSiemensVoIPChallenge(03‐06)– Almost‐reliable,real‐Qmetransport
• TheLiveTimeNetTVChallenge(08‐…)– FromOverlaystoClouds– UlQmateresiliency,automatedmonitoringandcontrol
• ThechallengesaheadFebruary16,2012 YairAmir 20
• Canwemaintaina“goodenough”phonecallqualityovertheInternet?
• Highqualitycallsdemandpredictableperformance– VoIPisinteracQve.Humansperceivedelaysat100ms– Thebest‐effortserviceofferedbytheInternetwasnotdesignedto
offeranyqualityguarantees– CommunicaQonsubjecttodynamicloss,delay,jiber,pathfailures
February16,2012 YairAmir 21
PSTN
50msnetworkdelay
• Localizedreal‐Qmerecoveryonoverlayhops– Retransmissionisabemptedonlyonce
• EachOverlaynodekeepsahistoryofthepacketsforwardedinthelast100ms– Whentheotherendofahopdetectsaloss,itrequestsa
retransmissionandmoveson
– IftheupstreamnodesQllhasthepacketinitshistory,itresendsit
• Notareliableprotocol– NoACKs.Noduplicates.Noblocking.
• Recoveryworksforhopsshorterthanabout30ms– Thisisok:overlaylinksareshort!
February16,2012 YairAmir 22€
loss ≈ 2 ⋅ p2
€
retr_ delay = 3 ⋅T + Δ
• Spinesoverlay–5linksof10mseach• 10VoIPstreamssendinginparallel• LossonmiddlelinkC‐D
February16,2012 YairAmir 23
50msnetworkdelay
• RouQngalgorithmthattakes
intoaccountretransmissions• WhichpathmaximizesthenumberofpacketsarrivingatnodeEinunder100ms?
• FindingthebestpathbycompuQnglossanddelaydistribuQononallthepossibleroutesisveryexpensive
• Weightmetricforlinksthatapproximatesthebestpath
February16,2012 YairAmir 24
€
Exp_ latency = (1− p) ⋅T + (p − 2 ⋅ p2) ⋅ (3 ⋅T + Δ) + 2 ⋅ p2 ⋅Tmax
• Localizedreal‐Qmerecoveryonoverlayhops– Retransmissionisabemptedonlyonce
• FlexiblerouQngmetricavoidscurrentlycongestedpaths– Costmetricbasedonmeasuredlatencyandlossrateofthelinks
– Linkcostequivalenttotheexpectedpacketlatencywhenretransmissionsareconsidered
February16,2012 25YairAmir
• TheOverlayNetworkParadigm• TheDARPANetworkingChallenge(99‐03)
– OverlayArchitecture– Low‐latencyreliabletransport
• TheSiemensVoIPChallenge(03‐06)– Almost‐reliable,real‐Qmetransport
• TheLiveTimeNetTVChallenge(08‐…)– FromOverlaystoClouds– UlQmateresiliency,automatedmonitoringandcontrol
• ThechallengesaheadFebruary16,2012 YairAmir 26
• CantheInternetbeanunderlyingnetworkforaliveTVservice?– Livechanneltransport(BusinesstoBusiness)– Thevirtualcablecompany(BusinesstoConsumer)– NextGeneraQonTV(InteracQvity)
• Requirements– Scalability:Highcapacityflows,manyany‐to‐manyflows
– Highavailabilityanduniformdelivery
• Technologytrends– Cheaplong‐haulaccessbandwidth– BroadbandInternetconnecQvitytothehome
– MulQ‐corecomputerarchitectureFebruary16,2012 27YairAmir
Theserviceproviderpointofview• AserviceratherthansoNwareorhardware• Controloverwhereoverlaynodesarelocated
• MulQplenetworkprovidersineachoverlaynode(SuperNodes)
• Guaranteedcapacitywithadmissioncontrol
• MonitoringandControl–nearautomaQon
February16,2012 28YairAmir
February16,2012 29YairAmir
February16,2012 30YairAmir
February16,2012 31YairAmir
February16,2012 32YairAmir
• Scalableoverlaynetworkarchitecture– Paralleloverlays
• Real‐Qmemonitoringandcontrol– Automated–takethehumanoutoftheloop
• ThreelevelsofprotecQon– Linklevel:real‐QmeprotocolforHD‐TV
– Overlaylevel:responsiveoverlayrouQng– Cloudlevel:NxWayfailoverforoverlayrouters
February16,2012 33YairAmir
Network packet loss on one link (assuming 66% burstiness)
Loss experienced by flows on the LTN Network
2% < 0.0003%
5% < 0.003%
10% < 0.03%
February16,2012 34YairAmir
Availablelink
SuperNode
35February16,2012 YairAmir
• UQlizesmulQpleTier1IPbackbones• OpQmizedoverlaypathsdetermineselectedlinks• AutomaQcallyandinstantaneouslyswitchtoabeberpath
Availablelink Selectedlink
SuperNode
36February16,2012 YairAmir
• UQlizesmulQpleTier1IPbackbones• OpQmizedoverlaypathsdetermineselectedlinks• AutomaQcallyandinstantaneouslyswitchtoabeberpath
Availablelink Selectedlink DeterioraQnglink
SuperNode
37February16,2012 YairAmir
• UQlizesmulQpleTier1IPbackbones• OpQmizedoverlaypathsdetermineselectedlinks• AutomaQcallyandinstantaneouslyswitchtoabeberpath
Availablelink Selectedlink DeterioraQnglink
SuperNode
38February16,2012 YairAmir
• UQlizesmulQpleTier1IPbackbones• OpQmizedoverlaypathsdetermineselectedlinks• AutomaQcallyandinstantaneouslyswitchtoabeberpath
Availablelink Selectedlink DeterioraQnglink
SuperNode
39February16,2012 YairAmir
• UQlizesmulQpleTier1IPbackbones• OpQmizedoverlaypathsdetermineselectedlinks• AutomaQcallyandinstantaneouslyswitchtoabeberpath
• UQlizesmulQpleTier1IPbackbones• OpQmizedoverlaypathsdetermineselectedlinks• AutomaQcallyandinstantaneouslyswitchtoabeberpath
Availablelink Selectedlink DeterioraQnglink
SuperNode
40February16,2012 YairAmir
• TheOverlayNetworkParadigm• TheDARPANetworkingChallenge(99‐03)
– OverlayArchitecture– Low‐latencyreliabletransport
• TheSiemensVoIPChallenge(03‐06)– Almost‐reliable,real‐Qmetransport
• TheLiveTimeNetTVChallenge(08‐…)– FromOverlaystoClouds– UlQmateresiliency,automatedmonitoringandcontrol
• ThechallengesaheadFebruary16,2012 YairAmir 41
• Resiliency‐allthewaytointrusiontolerance– Resilientclouds– CriQcalinfrastructure– SKYDA(SCADAintheSky)
• Timelinessandquality–noendtothat – RemotemanipulaQon
– Remotesurgery– Remotemusictraining?
February16,2012 YairAmir 42