Re: [trill] TRILL OAM Requirements: available paths

On 4/27/12 2:15 PM, Santosh Rajagopalan wrote:
> I agree, the wording needs to be tightened to say "all possible ecmp
> paths" rather than "all available paths". An operator working out R1-R2
> communication is interested in all possible paths data traffic could
> take rather than all theoretically available paths.

The issue is that there is nothing in the TRILL dataplane nor IS-IS 
control plane which helps the operator determine that some paths 
(through RB121 and RB112 in the example) are not possible; the ECMP 
input and hashing is a completely local matter on each RBridge.

It isn't useful to state a requirement around something that can't be 
determined.

    Erik

>
>
> From: Erik Nordmark <nordmark@acm.org>
> To: trill@ietf.org
> Date: 04/27/2012 09:53 AM
> Subject: [trill] TRILL OAM Requirements: available paths
> Sent by: trill-bounces@ietf.org
> ------------------------------------------------------------------------
>
>
>
>
> Two requirements use the wording "all available paths." and "all
> available ECMP paths.", respectively and I'm not sure I know what that
> means.
>
> Take the following example picture:
>
> R1
> / \
> / \
> / \
> / \
> / \
> RB11 RB12
> / \ / \
> / \ / \
> RB111 RB112 RB121 RB122
> \ \ / /
> \ \ / /
> \ \/ /
> ------R2------
>
> We can see that there are 4 paths with 3 hops to get from R1 to R2, with
> two layers of ECMP decisions. But how many available paths are there?
>
> Given that the RBridges make independent ECMP decisions, and how they do
> that is a local matter, it could be the case that for any packet, if R1
> picks the left link, then R11 would also pick the left link; if R1 picks
> the right link, then R12 also picks the right link.
> Thus there would be no packet (data or OAM) that would cause packets
> between R1 and R2 to pass via RB112 or RB121.
>
> Even if we ignore that as unlikely, there is still the fact that R1
> can't easily tell how many (shortest) paths there might be between it
> and R2. With single-level TRILL the LSDB could be used to tell, but if
> we ever go to multi-level that wouldn't be the case any more.
> And even if R1 knows that there are 4 potential paths, it can't format a
> data packet (or format the flow entropy in an OAM packet) in such a way
> that particular paths get chosen by downstream ECMP decisions.
>
> Thus for any notion of "all available paths" that make sense to me, we'd
> need an OAM approach that can explore the topology one hop at a time and
> at each hop explore all the possible ECMP choices. Using the entropy
> doesn't help with this (but the entropy is critical for following the
> path taken by a particular flow of data packets).
>
> Hence it would be good to clarify what the assumptions are behind this
> notion of all available paths.
>
> Thanks,
> Erik
>
> _______________________________________________
> trill mailing list
> trill@ietf.org
> https://www.ietf.org/mailman/listinfo/trill
>
>