The MPLS WG Archive

[Bora Akyol <akyol@pluris.com>]

Good

I think we are saying the same thing, but I made some clarifications to my
original email and want to see if you agree.

Thanks

Bora


Kireeti Kompella wrote:

> Hi,
>
> > > When filling in the ERO, there are three options:
> > >
> > > 1) Put in the router ID of the routers. This does not work for cases
> > > when routers have multiple links and for unnumbered interfaces.
>
> Technically this always works.  You lose control of exactly which
> interface to use, but that is a perfectly valid ERO.
>

This is not good when you have bandwidth guarantees across the multiple
interfaces. Of course if you have bonded interfaces like we do then this is a
non-issue.


>
> > > 2) Put in the egress interface of the routers on the path and finish it
> > > with the router ID of the final router. So that from router A to router
> > > B, we represent the hop as the IP address/interface index of the
> > > interface from A to B on router A.
>
> I interpret this as "remote IP address on each link", i.e., put B's
> address (or outgoing interface index from A's point of view) for the
> A->B link.
>

Not quite true. This means that I put in A's address on the A->B link. I believe
this is what you specified in the unnumbered draft as well. (Since from B's
perspective, A is the PHOP) For unnumbered interfaces, this means that I put the
ifindex of the outgoing link as seen from A's and then B's perspective.

See below:

A->B->C->D

the ERO then looks like: (Let ID(A,B) represent the ID of the outgoing link at
hop A towards B)

ID(A,B)
ID(B,C)
ID(C,D)
ID(D)

where ID(A,B) never actually makes it out of router A.

Do you agree?



>
> Note that "finishing with router ID of final router" is not needed.
>
> > > 3) Put in the ingress interface of the routers on the path starting with
> > > the router ID of ingress and finishing with the router ID of the egress
> > > LSRs. This is kind of opposite of (2).
>
> I read: "put the local address of each link, i.e., A's address for
> the A->B link".  In the unnumbered case, you are stuck -- you could
> somehow put B's outgoing interface id for the A-B link, but it would
> do A no good in finding the next hop.
>

Exactly true!

>
> Note that while either (2) or (3) works for point-to-point links,
> only (2) works for multipoint links.
>
> FWIW, there is yet another option: put both the local and remote
> interface addresses in the ERO.  Yes, the ERO size is doubled, but
> this works, and there are even some situations where this makes sense.
>
> > > I believe that (2) is the only option that works with unnumbered
> > > interfaces.
>
> >From the unnumbered draft:
>
> 6.1. Interpreting the Unnumbered Interface ID Subobject
>
>    The Interface ID is the outgoing interface identifier with respect to
>    the previous node in the path (i.e., the PHOP).
>
> You are saying the same thing -- use outgoing indices in the ERO.
> (Which I believe is Markus's point.)  In any case, we're in sync :-)
>
> Kireeti.