The MPLS WG Archive

[John Drake <jdrake@calient.net>]

exactly

-----Original Message-----
From: Yakov Rekhter [mailto:yakov@cisco.com]
Sent: Tuesday, July 11, 2000 11:29 AM
To: Lou Berger
Cc: Greg Waters; Bora Akyol; George Swallow; mpls@UU.NET
Subject: Re: Link Bundling 


Lou,

> At 08:35 AM 7/11/00 -0400, Greg Waters wrote:
> >At 07:48 PM 7/10/00 -0400, Lou Berger wrote:
> >>Greg,
> >>         You are correct, spreading a single LSP over multiple component

> >> links is not supported in the bundle draft.  I agree this could have 
> >> some value for best-effort LSPs.  (The draft was concerned with TE 
> >> LSPs.)  We could support such "spreading" by adding a special link
value 
> >> or bit to indicated that the label should be made available on all PSC 
> >> (packet) links of a bundle/bond.  (The spreading algorithm need not be 
> >> standardized for this case.)
> >>Lou
> >
> >I was talking about TE LSPs--in particular, TE LSPs whose capacity is 
> >greater than the fastest physical link.  Having that kind of LSP greatly 
> >simplifies the scaling and resiliency of TE.  You can widen the bundle at

> >any hop without having to thread new LSPs to use it.  You can, for 
> >example, increase the capacity of an existing LSP without regard to the 
> >elements of bundles that it traverses.
> 
> I may be wrong (since I of course can't speak for anyone else) but, I
don't 
> think any of us would argue that allowing LSPs to span multiple links is 
> worth investigating.  That said, I do feel that we should get the simple,
1 
> link/LSP, solution out there without having it wait for the harder to 
> standardize, n-links/LSP, solution.

For some types of links (e.g., links between OXCs) n-links/LSP solution
is just not feasible (due to the nature of the links). And we certainly
need to have a solution for such links.

Yakov.