Lecture11-Networks Interconnection ModelsMPLS.ppt

  • View

  • Download

Embed Size (px)


  • Chapter 10 Cc Cu trc Mng Tin tinCc M hnh kt ni Mng

  • Network InterconnectionServer network (Network 2) provides transport service to Client networks (Network 1 & Network 3)Control Plane Issues:Server network & client networks may use different technologiesWhat signaling is used and how are paths determined?ATM NetworkIP NetworkIP NetworkSONET NetworkOptical Network

  • End-to-End Protocol StacksExample: IP over ATMHosts run TCP/IPClient networks are IP networksServer network is ATM

  • Approaches to InterconnectionOverlay ModelIndependent control planesClient interacts with server network through User-Network Interface (UNI)Signal across UNI to request or release connectionsNo network state information passes from server network to client networkSecure & appropriate when networks run by different administrationsAddressing method in client & server networks differentNeed ARPClient & server networks can evolve independentlyPeer-to-Peer ModelSame control plane spans client & server networkClient network knows state of server networke.g. OSPF information shared among networksRSVP implemented in all networksClient network can make routing decisions involving server networkHigher efficiencySame addressing scheme in client and server networksNo need for address resolution protocolInterdependence makes evolution more difficult

  • Overlay Example: IP over ATM Multiprotocol over ATM (MPOA) uses overlay approachEdge Device (ED) interposed between IP net & ATM netED contains MPOA client (MPC) to set up & release VCsATM has MPOA servers (MPS) for IP-ATM address resolution & IP packet forwarding

  • EDEDMPS2MPC1MPC2MPS1MPS3Host1Host2Default pathClientnetworkClientnetworkShort-cut pathEdge deviceATM switchIP routerOverlay Example: IP over ATM First packets from Host 1 to Host 2 are routed using MPSsIngress ED monitors packet flowsWhen long-lived flow detected, MPD decides to set up VC Sends ARP request, which is routed along routed pathReply informs ingress ED of egress EDs ATM addressVC set up & subsequent packet use ATM shortcut

  • Routing Scalability in Overlay ModelRouters are interconnected with ATM VCs in full meshMany router adjacencies N2 for full meshRouting algorithm becomes unnecessarily complexMany message exchanges when topology changesRouting could be simplified if ATM nodes used IP routingMPLS addresses this problemATMnetwork

  • IPATMPHYxxxxxABCDClientIPClientIPServer networkPeer-to-Peer Example: IP + ATMNodes combine ATM switching & IP routingInitially packets are routed, hop by hopPackets flow along default VCs xWhen long-lived flow detected, node sets up shortcutClient establishes VC shortcut y1Node A establishes VC shortcut y2And so on

  • Chapter 10 Advanced Network ArchitecturesMPLS

  • What is MPLS?Multiprotocol Label Switching (MPLS)A set of protocols that enable MPLS networksPackets are assigned labels by edge routers (which perform longest-prefix match)Packets are forwarded along a Label-Switched Path (LSP) in the MPLS network using label switchingLSPs can be created over multiple layer-2 linksATM, Ethernet, PPP, frame relayLSPs can support multiple layer-3 protocolsIPv4, IPv6, and in othersIPIPLERLERLSRLSR

  • Why MPLS?Labels enable fast forwardingBut longest-prefix match is also fastCircuits are good (sometimes)Conventional IP routing selects one path, does not provide choice of routeLabel switching enables routing flexibilityTraffic engineering: establish separate paths to meet different performance requirements of aggregated traffic flowsVirtual Private Networks: establish tunnels between user nodes

  • Proposals Leading to MPLSIP Switching: IP+ATM proposed by IPSILONTraffic-driven label assignment: create & teardown shortcut paths according to flow activityCell-Switch Router: proposed by ToshibaTraffic-driven label assignmentTopology-driven label assignment: when node changes entries in IP routing table new ATM shortcuts are created & torn down Request-driven label assignment: signaling can request setting up of new labels to set up explicit pathsTag Switching: proposed by CiscoMultiprotocol tag or label: over multiple layer-2 technologiesLabel stacking: generalizes ATM 2-level hierarchyTopology-driven & request-driven label assignmentARIS (Aggregate Route-Based IP Switching): proposed by IBMLabel merging: optimization of label usage

  • Separation of Forwardng & ControlBefore MPLS: forwarding & control intertwinedTransition to CIDR (control) meant forwarding had to change to longest-prefix matchWith MPLS: forwarding & control are separateAll forwarding done with label switchingDifferent control schemes dictate creation of labels & label-switched pathsControl & forwarding can evolve independently

    All proposals leading to MPLS separate forwarding and control

  • Labels and PathsLabel-switched paths (LSPs) are unidirectionalLSPs can be:point-to-pointtree rooted in egress node corresponds to shortest paths leading to a destination egress router

  • Forwarding Equivalence Class FEC: set of packets that are forwarded in the same mannerOver the same path, with the same forwarding treatmentPackets in an FEC have same next-hop routerPackets in same FEC may have different network layer headerEach FEC requires a single entry in the forwarding tableCoarse Granularity FEC: packets for all networks whose destination address matches a given address prefixFine Granularity FEC: packets that belong to a particular application running between a pair of computers IP2IP2LERLERLSRLSRIP1IP1IP1IP2

  • VPI/VCIATM cellMPLS LabelsLabels can be encoded into VPI/VCI field of ATM headerShim header between layer 2 & layer 3 header (32 bits)20-bit label + 1-bit hierarchical stack field + 8-bit TTL3-bit experimental field (can be used to specity 8 DiffServ PHBs)

  • Label StackingMPLS allows multiple labels to be stackedIngress LSR performs label push (S=1 in label)Egress LSR performs label pop Intermediate LSRs can perform additional pushes & pops (S=0 in label) to create tunnels Above figure has tunnel between A & G; tunnel between B&FAll flows in a tunnel share the same outer MPLS label

  • Non-VC mergingVC mergingVC Merging Conserves LabelsAAL5 End-of-Packet bit can be used to reassemble packets

  • LSR 1LSR 2Label request for 10.5/16(10.5/16, 8)Label DistributionLabel Distribution Protocols distribute label bindings between LSRsupstreamdownstreamDownstream-on-Demand ModeLSR1 becomes aware LSR2 is next-hop in an FECLSR1 requests a label from LSR2 for given FECLSR2 checks that it has next-hop for FEC, responds with label

  • LSR 1LSR 2(10.5/16, 8)Label DistributionupstreamdownstreamDownstream Unsolicited ModeLSR2 becomes aware of a next hop for an FECLSR2 creates a label for the FEC and forwards it to LSR1LSR2 can use this label if it finds that LSR2 is next-hop for that FEC

  • Independent vs. Order Label Distribution ControlOrdered Label Distribution Control: LSR can distribute label ifIt is an egress LSRIt has received FEC-label binding for that FEC from its next hopIndependent Label Distribution Control: LSR independently binds FEC to label and distributes to its peersLERLERLSRLSR

  • Label Distribution ProtocolLabel Distribution Protocol (LDP), RFC 3036Topology-driven assignment (routes specified by routing protocol)Hello messages over UDPTCP connection & negotiation (session parameters & label distribution option, label ranges, valid timers)Message exchange (label request/mapping/withdraw)LSRLSR

  • MPLS Routing ScalabilityLSRs are visible to non-MPLS routersFewer router adjacencies Routing traffic & processing load reducedATMnetwork

  • RSVP-TEExtensions to RSVP for traffic-engineered LSPsRequest-driven label distribution to create explicit route LSPsSingle node (usually ingress) determines routeEnables traffic engineeringRSVP Path message includeslabel request object to request label bindingExplicit route object (ERO)RSVP Resv message includes label object

  • MPLS SurvivabilityIP routing recovers from faults in seconds to minutesSONET recovers in 50 msMPLS targets in-between path recovery timesBasic approaches:Restoration: slower, but less bandwidth overheadProtection: faster, but more protection bandwidthRepair methods:Global repair: node that performs recovery (usually ingress node) may be far from fault, depends on failure notification messageLocal repair: local node performs recovery (usually upstream from fault); does not require failure notification

  • MPLS RestorationNo protection bandwidth allocated prior to faultNew paths are established after a failure occursTraffic is rerouted onto the new paths

  • MPLS ProtectionProtection paths are setup as backups for working paths1+1: working path has dedicated protection path1:1: working path shares protection pathProtection paths selected so that they are disjoint from working pathFaster recovery than restoration

  • Generalized MPLSMPLS:Connection-oriented Leverages IP routing protocols, with TE extensions, to provide means for selecting good pathsProvides signaling for establishing pathsWith appropriate extensions, Generalized MPLS can provide the control plane for other networks:SONET networks that provide TDM connectionsWDM networks that provide end-to-end optical wavelength connectionOptical networks that provide end-to-end optical fiber path

  • Hierarchical LSPsGMPLS allows node with multiple switching technologies to be controlled by one control componentNotion of label generalized:TDM slot, WDM wavelength, optical fiber portLSP Hierarchy extended to generalized labelsMPLS LSP over SONET circuit o