RE: Optical Link Interface

[Kireeti Kompella <kireeti@juniper.net>]

Hi,

I wade into this discussion with trepidation.

I'll send out two mails.  This one is as a CCAMP co-chair, and
attempts to distill the issues so far.

This plane is the LMP-DWM vs. NTIP plane.  If you have other
travel plans, this would be an excellent time to disembark.

Both protocols satisfy the requirements.  Thanks to the authors
of the requirements draft to reach agreement on this!

The issues (so far) are:

a) LMP-WDM as an extension of LMP vs. NTIP as a new protocol.

   LMP-WDM is an extension of LMP, a reasonably mature protocol.
   NTIP is a new protocol.  These facts may mean it would take
   less time to standardize LMP-WDM than NTIP.  How relevant this
   is, I can't say.

b) Two protocols vs. one.

   NTIP and LMP-WDM are two protocols with one purpose.  I don't
   want two protocols with one purpose.  (As far as I understand,
   the ADs and the TA don't want two protocols with one purpose.)

   To make this crystal clear: LMP & NTIP is acceptable.  LMP &
   LMP-WDM is acceptable.  LMP-WDM & NTIP (with or without LMP)
   is not.

   "Let the market decide" is a final resort ... let's try not to
   go there.

c) The model: master-slave vs. peer.

   This is the key debate (in my opinion).  Is LMP-WDM a "natural
   extension of LMP", as Andre says; or is the master-slave model
   more appropriate in this context, as Osama says?

d) The transport protocol: TCP or raw over IP.

   Does the reliability of using TCP as transport offer a real
   advantage in this context?  Or do the complications of
   managing failover with TCP outweigh this benefit?

An aside: the issue of deployment keeps coming up.  The IETF does
not deploy protocols: it proposes standards.  Ideally, the proposed
standards meet requirements (from service providers!), which hopefully
indicates a willingness to deploy those proposals.  Deployment comes
into play when considering moving a proposed standard to standard
(and a standard to historic) -- but we are far from there.

If deployment was a criterion, the IETF would have killed CR-LDP a
long time ago, making the world a sadder, duller place, but making
my life (as implementor, draft writer and WG chair) much simpler.

Kireeti.