The MPLS WG Archive

Cell Relay Retreat>MPLS WG Archive>month:2005-Apr> msg00005



[Date Prev][Date Next][Thread Prev][Thread Next]  
  [Date Index][Thread Index][Author Index][Subject Index]

[mpls] LDP P2MP Extensions : loops ?

  • From: MORIN Thomas RD-CORE-LAN <thomas.morin@rd.francetelecom.com>
  • Date: Fri, 08 Apr 2005 17:21:19 +0200
  • Organization: France Telecom R&D CORE/CPN
  • X-OriginalArrivalTime: 08 Apr 2005 15:21:24.0547 (UTC)FILETIME=[A05CA930:01C53C4E]

Hi,

Two recent drafts have recently been submitted and propose extensions to
LDP to setup multipoint MPLS trees (P2MP and/or MP2MP). 

The question I have is about the impact of transient loops that may
occur because of e.g. IGP inconsistencies, or because of the delay
needed to update label switching tables.

As you know, IP multicast forwarding implies systematically doing an RPF
check against a packet source address, and dropping packets arriving on
an interface different from the one that would be used to reach the
source, thus effectively preventing exponential growth of the traffic on
a loop (always only one upstream interface).

With proposed LDP extensions for P2MP, its not hard to see how this
RPF-check is "emulated" : impose a break-before-make procedure (labe
withdraw THEN label map).

But for MP2MP, it seems to me that nothing like this can be done :
you'll have to accept traffic on many interfaces, and the label hides
the packet source so no RPF-like check can be done, except by looking
inside the labeled packet (!...).  Thus, it looks to me that
exponential-traffic loops can occur during IGP reconvergence in the
MP2MP case.

So I'd like to know if this problem was considered 
during the design of the MP2MP solution described in
draft-wijnands-mpls-ldp-mcast-ext-00 ?

-Thomas



_______________________________________________
mpls mailing list
mpls@lists.ietf.org
https://www1.ietf.org/mailman/listinfo/mpls