anintroductiontompls

anintroductiontompls内容摘要：

– Label bindings cannot flow from destination to source, but must be requested at source MPLS was initially designed to exploit the existence of ATM hardware and reduce the plexity of overlay works. But IP/MPLS with native ATM labels results in a large number of problems and plications. 31 Basic MPLS Control Plane MPLS control plane = IP control plane + label distribution Label distribution protocols are needed to (1) create label FEC bindings (2) distribute bindings to neighbors, (3) maintain consistent label swapping tables 32 Label Distribution: Option I “Piggyback” label information on top of existing IP routing protocol • Guarantees consistency of IP forwarding tables and MPLS label swapping tables • No “new” protocol required • Allows only traditional destinationbased, hopbyhop forwarding paths • Some IP routing protocols are not suitable • Need explicit binding of label to FEC • Link state protocols (OSPF, ISIS) are implicit, and so are not good piggyback candidates • Distance vector (RIP) and path vector (BGP) are good candidates. Example: BGP+ Good Points Bad Points 33 Label Distribution: Option II Create new label distribution protocol(s) • Compatible with “Link State” routing protocols • Not limited to destinationbased, hopbyhop forwarding paths • Additional plexity of new protocol and interactions with existing protocols • Transient inconsistencies between IP forwarding tables and MPLS label swapping tables Good Points Bad Points Examples: LDP (IETF) and TDP (Cisco proprietary) 34 The Control Plane IP IP Forwarding Table IP in IP out IP Label Swapping Table MPLS in MPLS out 77 data 23 data IP Routing Protocols + IP Routing Tables Label distribution protocols + Label Binding Tables Routing messages Label distribution messages 35 Label Distribution with BGP Carrying Label Information in BGP4 (1/2020) Associates a label (or label stack) with the BGP next hop. Uses multiprotocol features of BGP: RFC 2283. Multiprotocol Extensions for BGP4 So routes with labels are in a different address space than a vanilla routes (no labels) 36 BGP piggyback not required for simple iBGP optimization 417 IP 666 IP IP IP 233 IP AS 888 AS 444 BGP route Internal BGP Map traffic to the LSP that terminates at the egress router chosen by BGP A B Routers A and B do not need full routing tables. They only need IGP routes (and label bindings). 37 BGP piggyback allows Interdomain LSPs IP AS 888 AS 444 BGP route With label 99 Internal BGP with label 99 Use top of stack to get to egress router, bottom of stack for LSP in AS 444. IP 99 233 IP 99 666 IP 99 417 IP 99 38 MPLS tunnels can decrease size of core routing state • Core routers need only IGP routes and LSPs for IGP routes • Implies less route oscillation • Implies less memory • Implies less CPU usage BUT: still need route reflectors to avoid full mesh and/or to reduce BGP table size at border routers BUT: since your core routers do not have full tables you now have all of the MPLS problems associated with ICMP source unknown (TTL, MTU, traceroute …) Are these really problems? 39 Label Distribution Protocol (LDP) RFC 3036. LDP Specification. (1/2020) • Dynamic distribution of label binding information • Supports only vanilla IP hopbyhop paths • LSR discovery • Reliable transport with TCP • Incremental maintenance of label swapping tables (only deltas are exchanged) • Designed to be extensible with TypeLengthValue (TLV) coding of messages • Modes of behavior that are negotiated during session initialization • Label retention (liberal or conservative) • LSP control (ordered or independent) • Label assignment (unsolicited or ondemand) 40 LDP Message Categories • Discovery messages: used to announce and maintain the presence of an LSR in a work. • Session messages: used to establish, maintain, and terminate sessions between LDP peers. • Advertisement mess。