Espresso - SDN For Public Internet - IETF

Transcription

Espresso - SDN for Public InternetSam K. AldrinOn behalf of Google Technical Infrastructure and Google Cloud PlatformIETF, July 19th, 2017

Google NetworkMore than a collection of data centersFASTER (US, JP, TW) 2016SJC (JP, HK, SG) 2013Unity (US, JP) 2010Network fiberPoints of presence 100Google Global Cache edge nodes

Google Cloud RegionsAdding 11 new regions3Netherlands 2London 3 rtBelgiumN Virginia3S Carolina33Mumbai23#Current regions and number of zones#Future regions and number of zonesTokyoTaiwanSingaporeSão Paulo3Sydney

Ubiquitous Cloud.10x ScalingDatacenterCampus & MetroWANNext-gen disaggregation ofstorage, memory and computeCloud regions and campusexpansion driving DCinterconnectCloud replication andbandwidth intensive cloudservices (e.g., turnkey video,IoT)10x10x10xStep Function Disruptions: Bandwidth, Latency, Availability, Predictability

The Pillars of SDN @ GoogleB4AndromedaJupiterWANInterconnectNFV and networkvirtualizationDatacenterNetworking

B4: Google's Software Defined WANB4: [Jain et al, SIGCOMM 13]BwE: [Jain et al, SIGCOMM 15]

AndromedaGoogle Infrastructure ServicesVNET: 10.1.1/24VNET: 192.168.32/24Load BalancingDoSACLsVNET: 5.4/16VPNNFVToRToRToRToRInternal Network10.1.1/2410.1.2/2410.1.3/2410.1.4/24

Google Datacenter Network InnovationCapacityAnd hardware scale that we could not buyJupiterWatchtowerFirehose1.0Saturn4 Post1.3Pb/s clustersin 2013Firehose1.1Time8

The Pillars of SDN @ GoogleB4AndromedaJupiterWANInterconnectNFV and ernet?

The Pillars of SDN @ GoogleB4AndromedaJupiterEspressoWANInterconnectNFV and networkvirtualizationDatacenterNetworkingSDN for publicInternet

Espresso in ContextB4Jupiter Data CenterGoogle

Espresso in ContextPeering MetroB2B4Jupiter Data CenterGoogleGoogle

Espresso in ContextUserPeering MetroB2EspressoB4Jupiter Data CenterGoogleInternetGoogle

Espresso: Before and AfterRouterCloudCentric1.0ProtocolsLocal viewConnectivity firstCoarse fault recoveryEspressoSDNPeeringPer-metro and global viewApplication signalsReal-time optimization

Espresso Architecture OverviewEspressoMetroPeering FabricBGPspeakerLabel-switchedFabriceBGP PeeringExternal Peer

Espresso Architecture OverviewEspressoMetroPeering FabricHostBGPspeakerLabel-switchedFabriceBGP PeeringExternal PeerHostHostHostHostHostPacketProcessorLabeled packetsspecify egressHostHostHostHostHost

Espresso Architecture OverviewGlobal ControllerEspressoMetroApplication SignalsLocalControlPeering FabricHostBGPspeakerLabel-switchedFabriceBGP PeeringExternal PeerHostHostHostHostHostPacketProcessorLabeled packetsspecify egressHostHostHostHostHost

Availability is Paramount First things first: an insecure infrastructure is an unavailable infrastructureStability is more important than efficiencyNetwork management is criticalConfiguration is hardAutomation matters but can be counter to availability“Evolve or Die: High-Availability Design Principles Drawn from Google’s Network Infrastructure.”SIGCOMM 2016.

Next DecadeChallenges in NetworkingThe next wave of computing Serverless compute in Cloud 3.0 IoTTightly coupled, general purposedistributed computingIt’s time to put it all together Agile Scale JitterIsolationPerformance is great, but onlymeaningful with availability,manageability, and velocity

Thank You!Thank You!

WAN Cloud replication and bandwidth intensive cloud services (e.g., turnkey video, . Real-time optimization. Espresso Architecture Overview Label-switched Fabric BGP speaker External Peer Espresso Metro . Global Controller BGP speaker eBGP Peering External Peer Espresso Metro Application Signals Labeled packets specify egress