The MPLS WG Archive

["Eric Gray" <eric.gray@sandburst.com>]

Thierry/Sumit/Manoj,

    Consider the following simple analysis:

o    assume that a router has N interfaces for routing in a
        given routing domain
o    assume (for simplicity) it is not multipathing on any of
        these interfaces
o    then - for any specific FEC -
    o    it has exactly one output interface to a nexthop for
            that FEC
    o    it has N-1 interfaces over which it might receive a
            packet destined for that nexthop and FEC
o    and, if there is a route change that changes the nexthop,
        then one of the N-1 previously upstream peers MUST
        be the new nexthop.

Therefore, it does not take a strange case to produce a scenario
in which a router that was forwarding packets to another becomes
the next hop for those same packets.

--
Eric Gray

So far, it has been writ:

> HI,
> > hi!
> >      IMHO this is not correct as the LSR from which the label
> > mapping has beeen
> > learnt can never be the upstream LSR for the same mapping - liberal or
> > conservative mode notwithstanding.
>
> Even if the IGP completely change the route for the FEC?
>
> e.g.
>
> If the network is           -------
>                            /       \
>                      A----B----C----D
>                                        |    |
>                                \----E
>
>
> First, the Path from A to E may be A B C D E. In this case,
>                             -------
>                            /       \
>                      A-->-B-->-C-->-D
>                                        |   \|/
>                                \----E
> hence, C is upstream with respect to D : C uses the label distributed by D.
>
> If the path from A to E changed to A B D C E. In this case,
>                             --->---
>                            /       \
>                      A-->-B----C--<-D
>                                        \|/  |
>                                \----E
> here, D is upstream with respect to C : D uses the label distributed by C.
>
> Finally, in liberal mode, if C has distributed a label to D (in the first
> case, where D was the Next Hop for C), it can use it directly.
>
> This is a quite strange case but it seems to be ok.
>
> Thierry
>