The MPLS WG Archive

[Curtis Villamizar <curtis@workhorse.fictitious.org>]

In message <5.1.1.11.2.20031225095050.047d7ea8@imb.m.ecl.ntt.co.jp>, "YAMAMOTO,
 Susumu" writes:
> Hi all,
> 
> I have a question on node backup mechanism in Fast Reroute.
> 
> According to the draft, the protected LSPs going out of a PLR have
> different labels for each, then especially in case of link/node failure,
> the PLR have to sweep entire label table and find out the LSPs
> to push a label for bypass LSP on. This process seems to take
> time much more than 50ms or so, because typically with existing
> routers partial changes of label table would take more than 1sec
> from our experiences.
> 
> Do you have already any implementation along the draft satisfying
> the processing time requirement, or I might misunderstand about
> the specification of FRR ?
> 
> Thanks,
> -Susumu 


Implementations exist in which FRR recovery is well under 50 msec.
The key is to set up the backup insegment and outsegment hardware in
advace so that only the mapping of incoming label to insegment has to
be changed.  For fastest recovery, this has to be done by intelligent
ingress line cards in response to a message from the egress line card
indicating which outbound interface (or node behind it) went down.

The data plane liveness draft is intended to cover the case where
framing (link layer) stays up or where the link layer has no fast
means of detecting a fault (for example, GbE).

Curtis