The MPLS WG Archive

[Shahram Davari <Shahram_Davari@pmc-sierra.com>]

>         I think you are confused about how it works.  You 
>don't need to do 
>a trace
>before you can send an Echo request nor do you need to know the hash
>keys a priori.

I think you need to read the draft again. Here is what is says:

"  An ingress LSR periodically sends an MPLS
   traceroute message to determine whether there are multipaths for a
   given LSP.  If so, each hop will provide some information how each of
   its downstreams can be exercised.  The ingress can then send MPLS
   echo requests that exercise these paths.  If several transit LSRs
   have ECMP, the ingress may attempt to compose these to exercise all
   possible paths."



>
>>2) The fact that there are many TLVs defined, means processing the 
>>ping/traceroute messages
>>requires lots of processing power.
>
>         That is an incorrect assertion. Ask anyone who has implemented
>it.
>
>>3) It is not clear in the event of a failure how can the Pinged node 
>>process all the incoming echo or traceroute messages 
>simultaneously. In 
>>other words it seems that it is not scalable.
>
>         I think you need to be more specific (i.e.: give a 
>detailed example).
>Throwing around the "not scalable" flag doesn't cut it.

That was the example I gave. In the event of a failure many LSP sources
would detect a loss of connection and would initiate diagnostics. They all simultaneously
would run traceroute and possibly more ping to diagnose the failure. The LSRs therefore
may need to process hundreds or thousands of traceroute/ping messages simultaneously.



>
>>4) LSP ping is a bidirectional transaction and therefore 
>requires 2x BW 
>>compared to
>>a unidirectional transaction. So it is not BW efficient either.
>
>         So. MPLS is unidirectional inherently connection-less,
>thus LSP ping mirrors this behavior nicely.  There may not be
>a reverse path from any node to the one in question because
>no reverse path (via MPLS) may exist.  I think you are assuming
>that a bi-directional LSP exists, in which case I assert that you
>should test both.  In this way it is simple and elegant.

What I meant was that in LSP-ping it is the ingress that detects the
failure not the egress. And for ingress to detect the failure you need
to return the echo response while if egress was to detect the failure
you wouldn't need to consume twice more BW to continuously return echo
replies.

-Shahram

>
>         --Tom
>
>
>>Yours,
>>-Shahram
>
>
>http://www.elsevier-international.com/catalogue/title.cfm?ISBN=
>155860751X
>
>