The MPLS WG Archive

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

In message <200311172111.QAA06667@hank.bcentralhost.com>, mark seery writes:
> 
> 
> > Therefore it is not "one of the fundamental precepts of routing" that
> > each node understand the ECMP algorithms used by the other nodes.
> 
> Curtis, it is grossly misleading to restate what I said in that way. I find t
> hat a little disappointing. But I accept your point that there is an existenc
> e proof that different algorithms can work in the same network. We will see i
> f this holds true when multi-protocol traffic is routed between different ven
> dors as well, perhaps as you say it will.
> 
> Mark.


Multiprotocol traffic is already being routed in multivendor networks
with ECMP.  It isn't pretty, but looking for a valid IP header and
falling back to per label split works.

All that is required is that each node recognize that traffic which
contains identifiable microflows, doesn't split traffic that does not
contain identifiable microflows, and that for traffic containing
identifiable microflows any given microflow takes a single path.
The PW PID formalizes the means to determine the traffic type.

Curtis