Re: [mpls] [PWE3] Multiple RSVP paths

In message <5E893DB832F57341992548CDBB3331639843877BA7@EMBX01-HQ.jnpr.net>
John E Drake writes:
>  
> Hi,
>  
> A related ID:
>  
> http://www.ietf.org/id/draft-kompella-mpls-rsvp-ecmp-00.txt
>  
> Thanks,
>  
> John

John,

The use of the E-bit is not clear.  If the sub-LSP all have E=0, is
the load split proportional to the sub-LSP reserved bandwidth?  If E=1
are the reserved bandwidths of the component LSP all ignored?  What
happens if E=0 for some sub-LSP and E=1 for others?

Also in the figure:

                          ------------- M -----
                         /                      \
                        /           --- P --     \
                       /           /         \    \
                      A --- X --- Y --- Q --- T --- B
                             \     \              / |
                              \     --- R -------   |
                               \                    |
                                ------- S ----------

It is possible to set up link bundle Y-P-T and Y-Q-T as Y-T at PSC-4,
then bundle Y-P-T-B and Y-R-B as Y-B at PSC-3, then X-Y-B and X-S-B as
X-B at PSC-2, and A-M-B and A-X-B as A-B as PSC-1.  Done with one PSC
value to spare (the implied PSC=0).

This use of link bundling would yield the finest control over the
allocation of resources.  Each link bundle is the sum of the
components.

The problem with link bundling and a limitation of the draft in
question is that some LSP over this bundled LSP cannot specify that it
*needs* to take one component (to be MPLS-TP compliant, for example),
and allow other LSP to indicate that it is fine to load split it.

[  And now a word for our sponsor.  :-)  ]
See draft-villamizar-mpls-tp-multipath-00 for further requirements.

Curtis

> > -----Original Message-----
> > From: pwe3-bounces@ietf.org [mailto:pwe3-bounces@ietf.org] On Behalf Of
> > Curtis Villamizar
> > Sent: Tuesday, July 06, 2010 8:30 AM
> > To: stbryant@cisco.com
> > Cc: draft-ietf-pwe3-fat-pw@tools.ietf.org; pwe3
> > Subject: Re: [PWE3] Multiple RSVP paths
> > 
> > 
> > In message <4C31E719.1060402@cisco.com>
> > Stewart Bryant writes:
> > >
> > > In response to a comment received during last call I have added a new
> > > section 7.3 entitled : "Multiple RSVP-TE Paths" which says:
> > >
> > > In some networks it is desirable for a Label Edge Router (LER) to be
> > > able to load balance a PW across multiple RSVP-TE tunnels. The flow
> > > label mechanism described in this document may be used to provide the
> > > LER with the required flow information, and necessary entropy to provide
> > > this type of load balancing. Methods by which the LER is configured to
> > > apply this type of ECMP is outside the scope of this document.
> > >
> > > - Stewart
> > 
> > 
> > Stewart,
> > 
> > You could mention using a link bundle with the all ones component
> > link as a possibility but still leave further details out of scope and
> > not preclude other mechanisms that are local to the LER.
> > 
> > Curtis
> > _______________________________________________
> > pwe3 mailing list
> > pwe3@ietf.org
> > https://www.ietf.org/mailman/listinfo/pwe3