|
Hi Greg,
Hi Giovanni, thanks for the close read. Looks like you caught
some problems with the text. See below for comments.
Giovanni
Martinelli (giomarti) wrote:
Hi
Greg,
Sorry for the
delay in replying. I'm working on this topic since a while so yes, it's
interesting. Before going on specific issue I would have some
question/clarification regarding the draft itself.
* Within Abstract
and the following.
You don't talk about
Optical Cross Connects (OXC) is something missing or understated
somewhere?
-->Whoops. We were trying to find a more general term to include
both ROADM (usually a highly asymmetric fabric) and an OXC (a completely
symmetric fabric, e.g., any ingress to any egress), but we seemed to have gone
with using the ROADM terminology to include both cases. Talked with some
equipment makers that planned/make "switches" that seemed to incorporate both
so we made sure the model could deal with both sparse and dense potential
connectivity. Diego had some terminology ideas but lately his e-mails have
been bouncing back to me. Any suggestions are appreciated, but we are
including both ROADM and OXCs.
My doubt was coming from the ROADM definition in section 2 and picture
used later on the draft. At least in my understanding the OXC is a general
case for ROADM (sort of multi-degree ROADM) but ok, it's a matter of
terminology.
* Section 3.1
where you state:
"A fixed
mapping between the
GMPLS label space
and these ITU-T WDM grids as proposed in [Otani] "
Does it implies a
sort of network level label space? How relate with usual local label
significance?
--> This mapping gives a mapping between labels and
wavelengths/lambda, just like in the SONET/SDH case we mapped the ITU-T G.707
"S, U, K, L, M " identification of SDH time slots to a label format in RFC4606
and again this was done in RFC4328 to map G.709 digital wrapper time slot
identification into a technology specific label format. In RFC3471 for
lambda switching we just get a 32 bit integer with no meaning attached. Every
network and every node could potentially map labels to lambdas in a different
way. In [Otani] they are following the RFC4606 and RFC4328 lead and using the
ITU-T DWDM and CWDM lambda grid standards to give a fixed association between
labels and lambdas just like between labels and TDM time slots in the SDH/ODU
case.
This doesn't change the local significance of labels. In the
wavelength switched optical case that is influenced by the presence or absence
of wavelength converters.
Ok. Local significance but global semantic (as pointed out by
Adrian in a previous mail).
* Section 3.4
Wavelength Converters
"Current or
envisioned contexts for wavelength converters are : ..."
Could we think to a description/model for wavelength
converter that is technology agnostic? Simply
something like: full conversion capability, partial conversion
capability with some constrains, and may be
others. --> The difference, between the all
optical techniques and the OEO based techniques makes that difficult.
* Section 3.4.
the following:
"4. Wavelength
converters that are O-E-O based will have a restriction
based on the
modulation format and transmission speed"
Not clear to me the
type of restriction here when OEO happens... probably I'm missing what you
mean here.
--> For example a typical O-E-O based wavelength converter would be
build around a 3R regenerator with a tunable laser. A 3R regenerator cares
about the modulation type say NRZ or RZ (and which flavor), and the symbol
rate since its also doing retiming. An all optical wavelength converter will
be fairly independent of these issues (except when we look at impairment
factors). Hence the OEO wavelength is going to be more signal specific than
the all optical.
ok more clear now, although it would be nice having a general
model as you marked with TBD.
* Section 4.1 when
you talk about Lightpath temporal characteristics:
"Lightpath
connection duration has typically been thought of as
approximately
three time frames: "
and the following
you define: dynamics, pseudo-static, static.
Why there?s a need
of this classification? When you us Short/long is compared to
what?
--> In most of the research literature and in optimization practice
different techniques are typically used in the dynamic versus static (or
psuedo static cases). In MPLS there is minimum interference routing
optimization techniques for the dynamic case. For the static case I could
apply multi-commodity flow optimization techniques to a batch of
connections. In the RWA literature there is a similar
differentiation. Exactly what information could be sent to help PCE
differentiate I'm not sure. In the case of static, batch optimization we can
just use the existing concurrent optimization hooks in PCE. For an individual
lightpath request it seemed that it would be helpful to know how long the
connection would last so we'd know how much computational effort we might want
to put into optimize it.
ok, clear. I still have doubt about quantifiers but fine for the
moment.
Thanks,
Giovanni
minor typo on your mail below: point (c) rfc4328
(not 4238) right? --> Yes. The G.709
signaling extensions RFC.
Thanks,
Giovanni
Hi folks, I haven't seen too many comments
on our draft "Framework for GMPLS and PCE Control of Wavelength Switched
Optical Networks" ( http://www.ietf.org/internet-drafts/draft-bernstein-ccamp-wavelength-switched-01.txt).
So I figured I'd point out some potentially controversial issues that the
draft brings up.
(a) The draft brings up models for the following
WDM network elements:
- WDM links
- Optical transmitters
- Wavelength Converters and OEO regenerators
- ROADMs, FOADMs, optical splitters and combiners.
For items (3) and (4) we are taking the
modeling lead rather than some other SDO. And for ROADMs, in
particular, we going beyond the classic ITU-T "fabric" model (M.3100)
which has been the mainstay of any connection oriented switch (TDM, ATM,
MPLS).
(b) The draft brings up three (not one, not two, but three)
different computational models for RWA which can impact GMPLS and PCE
protocols:
- A single PCE computing both the path and wavelength
- Two distinct PCEs, where one computes the path, and a different PCE
computes the wavelength assignment
- A PCE computes the path and wavelength assignment is accomplished in
a distributed fashion via signaling (e.g., using label set objects)
Do we really need all three
models?
(c) G.709 includes the Optical Multiplex Section and
Optical Channels. RFC4238 was aimed at GMPLS extensions for
G.709 (Optical Transport Network) control. Weren't we finished
with all this optical stuff years ago?
I'd like to think the draft
answers some of these questions. I also think that network element
models and the process models are important enough to warrant this
separate framework document. Your opinions are
solicited.
Regards
Greg B.
--
===================================================
Dr Greg Bernstein, Grotto Networking (510) 573-2237
--
===================================================
Dr Greg Bernstein, Grotto Networking (510) 573-2237
|