The MPLS WG Archive

[Rahul Aggarwal <rahul@juniper.net>]

Hi Dave,

On Tue, 18 Nov 2003, David Allan wrote:

>
> Rahul:
>
> You are missing the point, if we do not know how ECMP is implemented, we
> cannot critique the solutions or participate in the design.

As has been articulated before on this thread, a MPLS OAM tool should be
able to work in the presence of ECMP, without prior knowledge of the hash
function. Lack of knowledge of the hash function does not preclude one
from designing an OAM solution or critiquing it. If certain tools have
constraints in an ECMP environment, the applicability of those tools
should be documented.

That said, if you really want to satisfy your curiosity and determine the
hash function used by a few major vendors, reverse engineering in the lab
is always possible :)

IMHO properly
> characterizing the dataplane is a pre-requisite to being able to design
> maintenance procedures. For example VCCV proposes using the router alert
> label as an alternative to the PW-CW **because** after much teasing out, the
> proposing vendor's ECMP implementation only hashes bottom label and payload
> and the RA label would be above the bottom label. Is that true of other
> vendors???? Is this solution acceptable only because everyone else would
> have to use the CW...(so there is an escape clause)?
>

The reason for having the router alert label is because a CW may not be
present or some h/w may not be able to support the CW solution. This has
the drawback of potentially breaking in the presence of ECMP and that is an
applicability of the router alert approach. We can clarify this in the
VCCV draft.

rahul

> IMHO the variations in ECMP that would be encountered by OAM are of
> practical interest to both protocol design and independent of generating a
> framework document. It is not an issue with the architecture, it is a
> pre-requisite to working on it.
>


> cheers
> Dave
>
> > -----Original Message-----
> > From: Rahul Aggarwal [mailto:rahul@juniper.net]
> > Sent: Monday, November 17, 2003 2:51 PM
> > To: Allan, David [CAR:NS00:EXCH]
> > Cc: 'tnadeau@cisco.com'; mpls@UU.NET
> > Subject: RE: on the mpls oam framework
> >
> >
> >
> > Hi Dave,
> >
> > >
> > > BTW, would be nice to get some other opinions, IMHO this is rather
> > > fundamental....
> > >
> >
> > To reiterate what has already been said on this thread. A
> > MPLS OAM framework document must a) Follow the MPLS
> > architecture and b) Discuss an OAM framework that concerns
> > *practical* problems seen in MPLS deployments.
> >
> > If folks have issues with MPLS architecture/usage, it should
> > be documented elsewhere, not in the OAM framework document.
> >
> > Regards,
> > rahul
> >
> >
> > > cheers
> > > Dave
> > >
> > >
> >
>