The MPLS WG Archive

["Gray, Eric" <egray@celoxnetworks.com>]

Curtis,

	You're applying the argument in the reverse of the
way I intended.  :-)

	In the same way that it is possible to argue that a
PLR knows where a back up path goes, so does an ECMP 
splitter.

	On a slightly different but related topic, note that 
- for the trace-route capability - the Downstream Mapping 
object makes it quite straight forward for the ECMP splitter 
to indicate to the head-end that there are multiple labels
and next-hops it might use for a specific FEC.  Too bad, is
it not, that this is not as true for the Ping capability.

Eric W. Gray
Systems Architect
Celox Networks, Inc.
egray@celoxnetworks.com
508 305 7214


> -----Original Message-----
> From: Curtis Villamizar [mailto:curtis@fictitious.org]
> Sent: Friday, December 13, 2002 2:54 PM
> To: Gray, Eric
> Cc: 'George Swallow'; curtis@fictitious.org; Shahram Davari; mpls@UU.NET
> Subject: Re: Last call on LSP Ping
> 
> 
> In message <1117F7D44159934FB116E36F4ABF221B0267EDA8@celox-ma1-
> ems1.celoxnetwor
> ks.com>, "Gray, Eric" writes:
> > And so, naturally, you could make this same argument for
> > how an ECMP splitter would test out all of the equal cost
> > paths...
> 
> 
> The ECMP split affects the primary path.  The traffic for a branch has
> to be injected upstream of the branch or you are not adequately
> testing it.  There is no way to send traffic from upstream of a detour
> to exercise a detour that is not in use.
> 
> Curtis
> 
> 
> > Eric W. Gray
> > Systems Architect
> > Celox Networks, Inc.
> > egray@celoxnetworks.com
> > 508 305 7214
> >
> >
> > > -----Original Message-----
> > > From: George Swallow [mailto:swallow@cisco.com]
> > > Sent: Friday, December 13, 2002 2:22 PM
> > > To: curtis@fictitious.org
> > > Cc: Shahram Davari; mpls@UU.NET; swallow@cisco.com
> > > Subject: Re: Last call on LSP Ping
> > >
> > > > Currently you'd have to do that from each PLR.  We don't provide a
> > > > means to tell the ingress of the existance of a detour (except the
> > > > local-protect-available bit in the signaling itself which one major
> > > > implementation wasn't using last I checked).  We also don't provide
> a
> > > > means to tell a specific PLR to exercise its detour and report back.
> > > >
> > > > Let's wait and hear from Kireeti or other authors whether they think
> > > > anything should be added or whether the intent is to require testing
> > > > initiated at each PLR.
> > >
> > > The PLR is responsible for the bypass.  It knows what sender-template
> > > it used, so it can form the proper FEC for the ping.  It can send a
> > > lsp-ping with the proper label stack.  So it is the natural place to
> > > do this.  If it fails, it should
> > >
> > > a) report to network management
> > > b) reset the backup in place bit
> > > c) look for another way of establishing a backup
> > > d) if successful on c) set the backup in place bit.
> > >
> > > You may want to put a hold-down on b) to avoid reporting to the
> > > head-end on a transient situation.
> > >
> > > I don't see what advantage you would gain by moving all this to the
> > > head-end.  I do see a lot of complications...
> > >
> > > ...George
> > >
> > >
> > > ==================================================================
> > > George Swallow       Cisco Systems                  (978) 497-8143
> > >                      250 Apollo Drive
> > >                      Chelmsford, Ma 01824
> >