RE: TE Requirements Draft - ELSP

["Sudhakar Ganti" <sganti@tropicnetworks.com>]

Francois,

> 3) using E-LSPs with traffic from multiple OAs each with their own SPF
> ======================================================================
> This is not allowed in recently posted REQTS draft.
> As Nabil/Sudhakar indicated, it is conceivable that some Service
> Providers
> might want to do that one day.
> But this require departure from the current DS-TE model and entails
> protocol extensions.
> I suggest we keep this option on the table for discussion for now
> but don't
> include it in the REQTS draft until we have established that a
> body of SPs
> actually want to do this and how it would actually be used (would we need
> just multiple bandwidth or also multiple affinity, multiple
> preemption,...).
> This would allow us to develop a faster simpler DSTE solution that
> addresses what we know SPs want to do today.
> We can always add this option at any time later when deemed useful and do
> the corresponding work then.
>
> Looking forward to hearing thoughts on this analysis/suggestion.
>
> Francois
>

First, I don't see a need to do per-OA SPF. Since it is a
single E-LSP carrying multiple-OAs, all of them are carried
on the same path. Therefore, the way I see the
SPF works is to take all the bandwidth parameters of OAs and come up
with a path that satisfies best the given constaints and optimization
criteria. The path computation may be a little involved. But this
is proprietary implementaion and standards need not specify it.
However, available bandwidth per-OA (or class) need to be
propagated which other OSPF drafts talked about.

Second point is that, when multiple OAs are carried on E-LSP, there
is no need to define multiple pre-emption attributes. The
pre-emption definitions apply to the whole LSP (and so to all the
OAs together) and not to each traffic class separately.
In fact, if we do multiple OA everything, what would
be the difference between multiple L-LSPs and one E-LSP carrying
multiple OAs?

So, other than extra per-OA signaling needed to account for proper
bandwidth accounting, there is no significant departure from the
current DS-TE model.

-Sudhakar