Re: [mpls] Reference Augmented Forwarding - MPLS RAF

[Robert Raszuk <[e-mail]>]

Hi Tarek,

Many thx for your review.

[TS]: we have discussed the idea of user-defined network actions (by
> distributing associated ancillary data out-of-band) in the design team. We
> also have been discussing if the scope of distribution needs to be
> domain-wide or LSP scope. We may need to prioritize this discussion.
>

OK good to know :).

[TS]: reading the above seem to suggest transit SR-MPLS (segment
endpoints) may need to copy the tuple after popping the top label.
This is not something that MPLS routers do today and may require
changes
>  to dataplane processing.
>
>
Yes it will require that change if someone is using SR-MPLS for TE (as
opposed as replacement for LDP).

[TS]: indeed, this would require doubling the state in (control and
> dataplane)… An SR network, would require potentially twice the number of
> labels multiplied by the number of algos/flex-algos deployed. Given that
> SPL is there, I don’t see why LSRs won’t be able to just peek below the top
> label and figure presence of RFI – which is closer to what the DT has been
> discussing.
>


Well not really doubling if you use domain wide labels. Only egress would
need to have two values.
And that does not mean that it needs to double.
In any given domain label bit 20 could indicate such difference with no
doubling.
But this is just food for thought .. really orthogonal to RAF.


The idea of LSR nodes copying the RFI SPL (and RAF) as packet traverses SR
> segments may be a bit of challenge to carrying ISD data specific to each SR
> segment. For example, section 2.4 “Delay Budgets for Low-Latency
> Applications” in draft-saad-mpls-miad-usecases.
>

Actually not. That data comes via mgmt/control plane custom to each segment
endpoint and single RAV is used.

Many thx,
Robert