SDN-B4原版完整课件.pptx

《SDN-B4原版完整课件.pptx》由会员分享，可在线阅读，更多相关《SDN-B4原版完整课件.pptx（42页珍藏版）》请在淘文阁 - 分享文档赚钱的网站上搜索。

1、B4:Experience with a Globally-Deployed Software Defined WANSushant Jain,Alok Kumar,Subhasree Mandal,Joon Ong,et al.ACM SIGCOMM,2013Presented by Ye Tian for Course CS05112Outline Background Design Traffic Engineering TE Protocol and OpenFlow EvaluationMotivation WAN links are typically provisioned to

2、 3040%average utilization.Deliver admirable reliability With over-provisioning and high-end routing gear Googles data center WAN We control the applications,servers,and the LANs all the way to the edge of the network Most bandwidth-intensive applications can adapt their transmission rate based on av

3、ailable capacity No more than a few dozen data center deployments,making central control of bandwidth feasible.Two WANs of Google User-facing network peers with and exchanges traffic with other Internet domains at Googles Edge Points of Presence(PoPs)http:/Two WANs of Google The B4 WAN Provides conn

4、ectivity among data centers First and largest SDN/OpenFlow deployments.Making Google one of the largest ISPs in the worldB4 Applications Applications(increasing volume,decreasing latency sensitivity)user data copies(e.g.,email,docs)to remote data center remote storage access for computation over inh

5、erently distributed data sources(e.g.,pagerank)large-scale data push synchronizing state across multiple data centers(social network).B4 Characteristics Elastic bandwidth demands can tolerate periodic failures with temporary bandwidth reductions.Moderate number of sites A few dozen End application c

6、ontrol Can control application priorities and control bursts,rather than overprovision Cost sensitivity traditional WAN(3040%overprovision)is prohibitively expensive.Outline Background Design Traffic Engineering TE Protocol and OpenFlow EvaluationOverview Each B4 site Switch hardware layer Site cont

7、roller layer Consist of Network Control Servers(NCS),hosting both OpenFlow controllers(OFC)and Network Control Applications(NCAs).A number of B4 site one connect to each cluster(datacenter)A global layer Overview Global layer Centralized application:SDN Gateway and a central TE server Each server cl

8、uster is a logical AS.Each cluster contains a set of BGP routers that peer with B4 switches at each WAN site.eBGP session between B4 site and cluster router Run B4 as a single AS providing transit among clusters running traditional BGP iBGP session between B4 sitesOverview Design choices Goal:Has to

9、 support existing distributed routing protocols Choice 1:Build one integrated,centralized service combining routing and traffic engineering Choice 2:Deploy routing and traffic engineering as independent services,with the standard routing service deployed initially and central TE subsequently deploye

10、d as an overlayOverview Each B4 site consists of multiple switches with potentially hundreds of individual ports linking to remote B4 sites.The TE abstracts each site into a single node with a single edge of given capacity to each remote site.All traffic crossing a site-to-site edge must be evenly d

11、istributed across all its constituent links.B4 routers employ a custom variant of ECMP hashing to achieve the necessary load balancing.Switch Design Build B4 switch from multiple merchant silicon switch chips 128-port 10GE switch with 24 1610GE chips.Contains an embedded Linux processor Run OpenFlow

12、 agent(OFA):translates OF messages into driver commands to set chip forwarding table entries.Network Control Functionality Most B4 functionality runs on NCS in the site controller layer co-located with the switch hardware NCS and switches share a dedicated out-of-band control-plane network.Paxos han

13、dles leader election for all control functionality.Use a modified version of Onix as OpenFlow Control(OFC).Maintain Network Information Base(NIB),such as topology OFAs maintain active connections to multiple OFCs,communication is active to only one OFC at a time and only a single OFC maintains state

14、 for a given set of switches.Routing Open source Quagga for BGP/ISIS on NCS Routing Application Proxy(RAP)as an SDN application to provide connectivity between Quagga and OF switches BGP/ISIS route updates routing-protocol packets between switches and Quagga interface updates from the switches to Qu

15、agga.RAP Quagga create tuntap interfaces corresponding to each physical switch port it manages.translates each RIB entry into two OpenFlow tables A Flow table which maps prefixes to entries into a ECMP Group table.Multiple flows can share entries in the ECMP Group Table.Proxy routing-protocol packet

16、s between the Quagga control plane Upon detecting a port state change,RAPd changes the netdev state for each interface change,which propagates to Quagga for routing protocol updates.Outline Background Design Traffic Engineering TE Protocol and OpenFlow EvaluationGeneral Architecture TE server operat

17、es on:Network Topology:a graph represents sites as vertices and site to site connectivity as edges.Flow Group(FG):a source site,dest site,QoS tuple A Tunnel(T)represents a site-level path A Tunnel Group(TG)maps FGs to a set of tunnels and corresponding weights.Output Tunnel Groups to SDN gateway,gat

18、eway forwards Tunnels and Flow Groups to OFC and install on Switch using Open FlowBandwidth Function Applications bandwidth function bandwidth allocation to an application given the flows relative priority(called fair share)FGs bandwidth function Piecewise linear additive composition of per-applicat

19、ion bandwidth functions.Example Bandwidth enforcer measures App1 demand:15Gbps A B App2 demand:5Gbps A B App3 demand:10Gbps A CFG1=App1+App2FG2=App3Tunnel Group Generation A greedy algorithm allocates edge capacity among FGs according to their bandwidth function such that all competing FGs on an edg

20、e either receive equal fair share or fully satisfy their demand.iterates by finding the bottleneck edge when filling all FGs together by increasing their fair share on their preferred tunnel.freeze all tunnels cross the bottleneck edge,move to the next preferred tunnel.Example Tx,y:yth most preferre

21、d tunnel for FGx.T1,1=A B,T2,1=A C,at share 0.90.A B becomes bottleneck T1,2=A C B,at share 3.33,A C becomes bottleneckExample T1,3=A D C B,T2,2=A D C,at share 10,A D and D C become bottleneck.FG2 is allocated a fair share of 10 while FG1 is allocated infinite fair share as its demand is fully satis

22、fiedTunnel Group Quantization Example split the above allocation in multiples of 0.5 FG2,(0.3,0.7)(0.0,1.0)FG1,(0.5,0.4,0.1)(0.5,0.5,0.0)Outline Background Design Traffic Engineering TE Protocol and OpenFlow EvaluationTE State and OpenFlow A switch maps packets to an FG when their destination IP add

23、ress matches the prefixes associated with the FG.Applications are under control Each incoming packet hashes to one of the Tunnels associated with the TG in the desired ratio.Source site switches encapsulate the packet with an outer IP header whose destination IP address uniquely identifies the tunne

24、l.The outer destination-IP address is a tunnel-ID rather than an actual destination.ExampleExample Half the flows are encapsulated with outer IP src/dest IP addresses 2.0.0.1,4.0.0.1 and forwarded along the shortest path The remaining flows are encapsulated with the label 2.0.0.1,3.0.0.1 and forward

25、ed through a transit site.After decapsulation,the switch forwards to the destination based on the inner packet header,using Longest Prefi x Match(LPM)entries(from BGP)Composing Routing with TE Routing/BGP populates the LPM table with appropriate entries TE uses the Access Control List(ACL)table to s

26、et its desired forwarding behavior.Incoming packets match against both tables in parallel.ACL rules take strict precedence over LPM entries.ExampleExample The LPM entry indicates that the packet should be forwarded through output port 2 without tunneling.ACL entry takes precedence and indexes into a

27、 third table,the Multipath Table,at index 0 with 2 entries.in parallel,the switch hashes the packet header contents,modulo the number of entries output by the ACL entry.i.e.,ECMP hashing distributing flows destined to 9.0.0.0/24 evenly between two tunnels.Coordinating TE States Across Sites Translat

28、e TE optimization output to a per-site Traffic Engineering Database(TED)key-value datastore for global Tunnels,Tunnel Groups,and Flow Groups.Compute per-site TED based on TGs,FGs,and Tunnels output by the TE algorithm.a single TE operation(TE op)can add/delete/modify exactly one TED entry The OFC co

29、nverts the TE op to flow-programming instructions at all devices in that site.Example Dependence Configuretunnel beforeTG and FG Cannot delete tunnel before removingaffected entriesOutline Background Design Traffic Engineering TE Protocol and OpenFlow EvaluationDeployment and EvolutionB4 traffic gro

30、wthDeployment and Evolution Topology change Change in capacity,286/day Add/remove edge on TE servers topology view,7/day links are susceptible to frequent port flaps and benefit from dynamic centralized management.TE Ops PerformanceTE Ops PerformanceImpact of Failures the failure of a transit router

31、 neighboring to an encap router requires a much longer convergence time Need to update all influenced entries in multipath table for tunnels traversing the failed router When TE fails,fall back to BGPTE Algorithm Evaluation(a)assume a 1/64 path-split quantum,to focus on sensitivity to the number of

32、available paths.(b)fix the maximum number of paths at 4,to show the impact of path-split quantum Link Utilization and Hashing 100%utilization ratio of high priority to low priority packets,and packet-drop fractions for each priority.Review Why Google wants to build a WAN?B4 networks system architecture.Each site Global layer TE over BGP TE algorithm Forwarding rules