Re: PADS BOF minutes?

[sven.van_den_bosch@alcatel.be]

Hi Nary, Ruediger,

I am sorry to jump into your discussion so late (been out of the office for
a few days). I would just like to offer my view on the nsis/pads
relationship.
1. NSIS is chartered to develop a generic signalling protocol that is based
on the lessons learned from RSVP. A design decision was made to distinguish
two separate layers in this protocol: NTLP (transport, focussing on getting
messages where they need to go) and NSLP (signalling, client or NSIS
'application' layer that implements the signalling client). NSIS has
decided to focus (at least initially) on a path-coupled (NTLP) approach,
where the path of the signalling messages is determined by 'normal layer 3
routing'. The PADS community feels that we should also consider the
path-decoupled case, mainly for two reasons:
- data and signalling path can not be guaranteed to coincide in all cases
- deployment is facilitated with path-decoupled (NTLP) signalling since in
that case only a limited amount of boxes have to be made NSIS (or PADS)
-aware: note that this is *not* restricted to ingress/egress like you
suggest. It is more likely a selected set of boxes along the data path,
e.g. one per AS. Note also that the NSLP should in principle be the same.
I believe that there is a real opportunity to come up with a really generic
protocol that can support both cases, provided we separate discovery from
message forwarding (like suggested in some proposals). There are several
other good reasons to do this, including security and flow control (on the
signalling entities).

Hope this helps,
Sven





"Thanh Tra LUU" <luu@enst.fr>@psg.com on 23/04/2003 13:34:21

Sent by:    owner-pads@psg.com


To:    "Geib, Ruediger" <Ruediger.Geib@t-systems.com>
cc:    <pads@psg.com>
Subject:    Re: PADS BOF minutes?


Hello Rüdiger,

|>Yes. In general a signaling unit must be signaling aware.
> Otherwise it is no signaling unit. What's your point?

Yes, it's sure.

But I mean that : do we need the processing of all entities that signaling
message cross over (like RSVP) or we only need the processing of the first
and the last point ( and some intermediate entites but not all ).

Do you mean that we can use NSIS protocol transparently (without knowing
it)
and we don't care how it is, we are only interested in the first point and
the last point of signaling ( and some intermediate entites but not all ),
there is no state on all of entities the signaling messages cross over ?


Regards.

Nary Tra

PhD Student
ENST, Paris, France
Tel : (0033) (0) 1 45 81 71 06