The MPLS WG Archive

["Bernstein, Greg" <GregB@ciena.com>]

Folks, been following some of the discussions of network abstractions,
and various models.  The Shared Risk Link Group TLV of Kompella, et.
al., <draft-kompella-mpls-optical-00.txt> is one such abstraction of
lower layer properties and provides some interesting examples, and
questions. Also looking at both optical and circuit switched examples
gives more interesting layering issues. 

As an example suppose we are given a non-dynamic WDM layer
composed of spans and optical add/drop multiplexers (OADMs), i.e., no
dynamic switching but where I add/drop wavelengths or bands of
wavelengths onto fibers. For each general span, i.e., a chain of WDM
links where no lambdas are added or dropped, I can assign a 32 bit "span
ID".  Every branch point where lambdas are added or dropped can be
defined as a new general span (my DWDM friends may not like my loose
terminology here) and would be assigned a new 32 bit "span ID". Given
that I have a SONET line (the layer about the SONET section layer) that
traverses this WDM system on its way from its source to destination, I
could assign the shared risk link group (SRLG) TLV to this line by using
the list of "span IDs" traversed (note that due to the OADMs a single
number may not suffice).  So I've just used WDM layer info to set the
properties of a SONET layer link (and this information can be used for
diverse routing calculations for example).

This brings a number of questions to mind.
(a) Okay, I've got friends at the WDM layer who can get me this info,
who else would really want this information?  There are no active
switching devices down there to control.  A similar situation comes up
in applying MPLS more generally to the circuit switched world where much
use is made of "fixed" multiplexers.  See
draft-mannie-mpls-sdh-control-00.txt
for examples of the intricate SDH hierarchy.
(b) Currently the SRLG of Kompella et. al. is an unordered list of 32
bit numbers. If I made this an ordered list in the above application I
would get more topology information concerning the SONET line, basically
an abstraction of the WDM path that it takes. Is there a reason why we
want this unordered vs. ordered?
(c) Being on the same fiber is just one example of the SRLG TLV, we
could have fibers in the conduits or conduits in the same right of way.
Some of this physical plant information would probably need to be
configured.  Given that one can't always find a diverse path based on
the SRLG lists of the various lines available (example routing SONET
paths over SONET line with these SRLG lists) it would be nice to
introduce some structure into the 32 bit SRLG number so we could in some
way optimize (or at least make better decisions) the diversity of the
paths. For example, fiber diverse is most important, conduit of second
most importance, etc...
(d) So now we go an throw some optical switches into the mix. Now how
much info needs to to distributed about the optical layer topology (lots of
analog info too?). And who should set up the routes (my nice smart
SONET level switch right? but what does he know of power budgets etc...).
How much information should the optical switch/DWDM system know about
the SONET layer? At the OIF we've be talking about at least end system
discovery via J0 snooping. How much does he really want to know? Similar
issues would arise in application of MPLS to the circuit switched
hierarchy, e.g., would a SONET switch really want to know about DS0
level detail?
(e) Seems like a good approach to hierarchy and info sharing across
layers is needed... Some of the various documents out there have some of
the pieces... Adjacencies, SRLG, bundling, etc.. But not quite the full
story yet.  For example, what if a node really doesn't want to be
bothered with higer layer routing information? 
Note that sometimes its good to look at the circuit
switched examples or the IP and circuit switching examples rather than
IP and ATM examples (more hierarchy, very different worlds from the
routing point of view, i.e., packet and switched rather than packet and
packet). Also, can we use the some of these natural hierarchical
switching boundaries to allow for easier configuration of OSPF areas
or something like them to allow the networks to scale up in size (think
about the circuit switched world and all those T1 and E1 lines that
could be configured and managed better...)

Greg B.
****************************************************************
Dr. Greg M. Bernstein, Senior Scientist Ciena Core Switching Division