RE: Optical Link Interface

[Kireeti Kompella <kireeti@juniper.net>]

Hi,

This email is as a router vendor/implementor.

The issues (so far) are:

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

   Subject to (c) below, I much prefer having to implement LMP
   (which I will have to) and add on the features needed for
   the OLI, rather than implement two different protocols.  The
   rationale is the same as was used when deciding to extend
   link state IGPs with TE information rather than do a new TE
   link state flooding protocol, or to do GMPLS as an extension
   of MPLS rather than develop a new optical signaling protocol.

b) Two protocols vs. one.

   One.

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

   I'm not qualified to speak here; this is the domain of the
   optical gurus.  However, I'll make two tangential points.  The
   first is that the answer seems to depend on the architecture
   of the component systems, which probably indicates that the
   protocol should do both.  The second is a piece of generic
   wisdom (?) that it is generally easier to add bells onto a
   peer-oriented protocol to deal with the master-slave case
   than to add whistles to a master-slave protocol to make it
   work in the peer case.

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

   It is my understanding that the PXC-WDM link is usually one IP
   hop; it seems that TCP is an overkill in that case.  Also,
   building failover capability for TCP-based protocols is rather
   more challenging.

   (Side note: both protocols can be made to work either way, so
   this is not so much of an NTIP-LMP_WDM issue as much as "what
   transport should OLI use?")

Kireeti.