Requirements for P2MP Extensions to LDP draftlerouxmplsmpldpreqs01.txt - PowerPoint PPT Presentation

1 / 13
About This Presentation
Title:

Requirements for P2MP Extensions to LDP draftlerouxmplsmpldpreqs01.txt

Description:

... with some common and some distinct capabilities and properties ... An upstream label allocation mode sounds relevant to ease such negotiation. Label spaces: ... – PowerPoint PPT presentation

Number of Views:63
Avg rating:3.0/5.0
Slides: 14
Provided by: jlle
Category:

less

Transcript and Presenter's Notes

Title: Requirements for P2MP Extensions to LDP draftlerouxmplsmpldpreqs01.txt


1
Requirements for P2MP
Extensions to LDP
draft-leroux-mpls-mp-ldp-reqs-01.txt


Jean-Louis Le Roux
(France Telecom)
Thomas Morin (France Telecom)
Vincent Parfait (Equant)
Luyuan Fang (ATT)
Lei Wang (Telenor)
Yuji Kamite (NTT Communications)
Shane Amante (Level 3 Communications)
  • IETF 63, Paris, MPLS WG, 08/01/2005

2
Does this ID fit in with the MPLS Charter??
  • This draft lists a set of functional requirements
    for P2MP extensions to LDP
  • From MPLS WG Charter
  • "procedures and protocols for multicast protocol
    extensions for point-to
  • multipoint TE, including soft-preemption"
  • What does it mean?
  • Some charter rewording would be highly benefic ?
  • Proposal "Procedures and protocols for
    point-to-multipoint MPLS, including Traffic
    Engineering"
  • Soft preemption is actually a separate item
  • Hence, can we say that P2MP LDP is part of the
    charter?
  • Strictly speaking NO
  • But after this cosmetic rewording YES ?

3
Problem Statement 1/2
  • LDP largely deployed for setting up unicast MPLS
    LSPs
  • Point-to-point MPLS traffic delivery
  • Major applications PE-to-PE tunneling in MPLS
    L3/L2 VPN networks
  • Emerging requirements for supporting multicast
    traffic delivery in these MPLS VPN networks (see
    ongoing work in L3/L2VPN WGs)
  • Use of MPLS tunneling for delivery of multicast
    traffic in these MPLS VPN networks is a valuable
    approach
  • Consistency with unicast traffic delivery
  • Take full advantage of MPLS infrastructure
  • Could rely on a group of unicast LSPs, with
    traffic replication on the Ingress LSR
  • But this is sub-optimal and may lead to
    significant bandwidth wasting

4
Problem Statement 2/2
  • Hence new mechanisms are required for setting up
    P2MP LSPs
  • P2MP LSP One Ingress LSR and one or more Egress
    LSRs, with MPLS traffic replication at some
    Branch LSRs
  • as already well pointed out in draft-ietf-mpls-p2m
    p-sig-requirements
  • Several possible approaches for setting up P2MP
    LSPs, with some common and some distinct
    capabilities and properties
  • RSVP-TE, LDP, or PIM extensions
  • Important work already done for P2MP TE LSPs
    signaling
  • See draft-ietf-mpls-rsvp-te-p2mp
  • For LDP-based MPLS networks without TE needs, a
    relevant approach would consist of relying on
    simple LDP extensions for setting up P2MP LSPs
  • Provided of course, that this does not introduce
    complexity to such a degree that it would
    diminish the benefits of reusing LDP ?

5
ID Objectives / Non Objectives
  • This document focuses on the LDP approach for
    setting up P2MP LSPs
  • List of detailed requirements for P2MP extensions
    to LDP
  • Some are common with the P2MP TE Reqs ID
    (draft-ietf-mpls-p2mp-sig-requirements)
  • To be used as guidelines when specifying LDP
    extensions
  • Non objectives
  • Generic requirements for P2MP MPLS
  • Comparison of various P2MP MPLS approaches

6
Illustration
PE
7
Requirements summary 1/4
  • P2MP LSP
  • Same data plane aspects (Branch LSR) as defined
    in the P2MP TE req ID
  • A P2MP LDP LSP MUST be identified by a constant
    and unique identifier
  • Need to define a FEC that is suitable for P2MP
    forwarding
  • P2MP LSP Setup/Teardown/Modification
  • It is RECOMMENDED to follow a leaf initiated LSP
    setup/teardown approach, so as to scale well with
    a large number of leaves
  • The way leaf LSRs discover LSP identifiers
    depends on the application and is out of the
    scope of P2MP LDP
  • MUST support dynamic addition/removal of Leaf
    LSRs
  • It is RECOMMENDED that this imply processing only
    on the path from the Branch LSR to the
    added/removed Leaf LSR

8
Requirements summary 2/4
  • Data Duplication
  • SHOULD be avoided and limited to rare transitory
    conditions
  • Routing Loops
  • Routing loops that may trigger non-localized
    exponential growth of traffic MUST be avoided
  • Any loop avoidance mechanism MUST respect
    scalability requirements
  • P2MP LDP Routing
  • MUST support hop-by-hop routing
  • SHOULD rely upon RIB information
  • P2MP LDP MUST support LSP re-routing
  • On a better path gt Packet loss SHOULD be
    avoided, potential data duplication MUST be
    minimized
  • Upon network failure gt Recovery time SHOULD be
    minimized, need for a mechanism to avoid LSP
    flapping
  • For planned maintenancegt SHOULD avoid packet
    loss, potential data duplication MUST be
    minimized

9
Requirements summary 3/4
  • P2MP LDP on LAN interfaces
  • MUST allow to send a single copy of the data to
    reach a set of downstream LSRs on a LAN segment
  • This requires that the same label be negotiated
    with all downstream LSRs
  • An upstream label allocation mode sounds relevant
    to ease such negotiation
  • Label spaces
  • MAY be shared or dedicated (gtDedicated implies
    separate LDP sessions)
  • Note that upstream label allocation requires
    dedicated label spaces (context specific label
    space)
  • P2MP LDP MUST be equally applicable to IPv4 and
    IPv6
  • Control and data plane

10
Requirements summary 4/4
  • P2MP LDP MUST allow setting up Multi-Area LSPs
  • Without any impact on IGP hierarchy
  • P2MP OAM
  • Required enhancements to the LDP MIB module
  • This may yield a new MIB module inherited from
    the LDP MIB
  • Precise requirements are already or will be
    addressed in the P2MP OAM ID
  • Graceful Restart/Fault Recovery
  • LDP GR and FT has to be enhanced to support P2MP
    LSPs
  • Scalability MUST be designed to scale well with
    an increase of Leaf LSRs, Branch LSRs, LSPs
    per LSR
  • Backward compatibility
  • MUST interoperate seamlessly with and inherit its
    capability set from unicast LDP
  • MUST not impede operations of unicast LSPs
  • SHOULD allow setting up P2MP LSPs among legacy
    non branch transit LSRs

11
Open Issues 1/2
  • Shared trees
  • Shared trees would be useful for traffic delivery
    between a group of N PE acting indifferently as
    Ingress or Egress.
  • This would avoid having N P2MP LSPs rooted at
    each PE, and would reduce the amount of LDP state
    to be maintained on an LSR
  • Various approaches to build such shared trees
  • MPLS level gt Multipoint-to-Multipoint LSPs
    (MP2MP LSPs)gt Need for specific LDP procedures
  • Application level gt Unicast LSPs to a root PE
    and one P2MP LSP from the root PE to each other
    PE gt See draft-ietf-l3vpn-2547bis-mcast-00.txt
  • Is there a need for shared trees?
  • Ongoing simulation effort to quantify the real
    gain of shared trees in operational MPLS networks
  • If yes, should it rely on application level
    procedures or on LDP procedures (MP2MP LSPs) or
    both?
  • MP2MP LSP Support is currently a MAY but this
    could evolve

12
Open Issues 2/2
  • Need for a survey of the expected order of
    magnitude of leaf LSRs, Number of LSPs per
    LSR, Leaf activity
  • Some parameters will be deduced form the
    Multicast VPN survey
  • Service Providers' feedback would be highly
    useful

13
Next Steps
  • Interest for this work ?
  • Need for WG feedback, particularly on open issues
  • WG doc ?
Write a Comment
User Comments (0)
About PowerShow.com