The MPLS WG Archive

[Ben Black <ben@layer8.net>]

On Wed, Dec 13, 2000 at 08:37:44PM -0500, Krishnan, Vijay G. wrote:
> Ben,
> 
> The LSR can calculate the MTU based on the number of labels it processes,
> but there could be more labels in stack.
> 
>  ingressLSR ------>LSR1------>LSR2------>LSR3------->LSR4------>egressLSR
>              GigE        POS        POS        POS
>                      |-----------Tunnel 1 ------------|
>                                 |---T2----|
> 
> Consider a case where the LSP goes through multiple (hierarchical) LSP
> tunnels. One from LSR1 to LSR4 and another from LSR2 to LSR3. Each LSR would
> know only that it is going to push  a label and wouldn't know how many
> labels the incoming packet has. It may process only the top two. LSR2
> calculates its MTU as 9K-2labels. But the packets through LSR2 to LSR3 would
> have three labels in stack (due to Tunnel 2). So packets fragmented on LSR1
> may need fragmentation at LSR2 also.
> 

This indicates the need for signalling of the MTU to head LSRs, which is what
I am asking be added to LDP.  Upon establishment of Tunnel 1 (or Tunnel 2)
a message could be generated back to head end LSRs for any LSPs crossing
the tunnels notifying them that the MTU had changed (this is not the same
as simply determining the MTU for an LSP at the time it is created).

> Just a thought... Can we have a field in the PATH message for label stack
> depth initialised to 1. It would be incremented by the LSRs when it is
> switched to LSP tunnels (label push). Maybe there are much better ways for
> finding the stack depth.
> 

An LDP Query, perhaps?


ben