[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
RE: Moving right along ...
Hi Maarten and Zhi,
I have the feeling that you are inventing a new model with a new protocol.
This is not how GMPLS works. I understand your frustrations and I have the
feeling that your ideas should be addressed somewhere else rather than
trying to change a model that we are finalizing and that we got from the
MPLS model.
There are maybe 1000 ways of designing a control plane for transmission
networks. I would suggest that you first develop your own model and then
that we compare the two.
Please, don't forget that we have here to be constructive in terms of GMPLS.
We have final discussions about the core GMPLS signaling, this is not the
beginning but the end of an (intermediate) phase.
Ok, GMPLS doesn't do everything. It doesn't do coffee and doesn't clean your
house :-) We know it. Personally, I don't believe in the ultimate model and
protocol that will do everything and that will take 10 years to be
specified, with abstract protocols, meta-models, etc.
By the way, personally I don't see any benefit of defining these
meta-models, abstract models and abstract protocols. It is hard to
understand for non super-expert and it is a complete loss of time for the
readers. I prefer practical solutions.
I am also seeing discussions that we already have many times in the past. I
don't see any benefit of having them again, and again, and again. If we
continue like this we will all implement a tool to answer automatically with
our previous e-mails :-)
Ok, there is a small set of people that don't like the way that GMPLS is
designed and defined (always the same 4 or 5 people by the way) - that's
100% your right guys and I respect it. In comparison with the hundreds
(thousands ?) of people that are working on GMPLS and implementing it, this
is a great achievement !
So, I ask you the question: what do you want to achieve personally at this
stage ? Don't you think that sometimes it is needed to finish one step,
before going to the next one ?
Kind regards,
Eric
-----Original Message-----
From: Maarten Vissers [mailto:mvissers@lucent.com]
Sent: Monday, October 22, 2001 11:07 PM
To: Yanhe Fan
Cc: 'Zhi-Wei Lin'; John Drake; Kireeti Kompella; ccamp@ops.ietf.org
Subject: Re: Moving right along ...
Yanhe,
The situation Zhi describes is one that will be very typical in the optical
(SDH/SONET) network in the coming years. We will most likely see islands of
switched subnetworks interconnected via the existing SDH/SONET non-switched
networks. By means of management actions, links (GMPLS: link bundles) will
be
setup between an edge node in one switched subnetwork and an other edge node
in
the next switched subnetwork. In G.805 terminology, these links are so
called
serial compound links.
The switched subnetworks will treat these serial compound links as normal
links,
"ignoring" the intermediate non-switched nodes they traverse. These links
will
contain a fixed set of link connections, which are identified by their
respective endpoints at nodes A and Z (i.e. SNPa - SNPz).
The server signal that carries the e.g. 5 STS1 link connections in the link
at A
can be an OC12, whereas this can be an OC48 at Z.
During discovery or by means of provisioning both A and Z nodes get aware of
the
5 sTS1 link connections as tuples of e.g.
SNPa SNPz
-----------------------------------
- port Ai/trib T1 - port Zj/trib Ta (STS1 link connection #1)
- port Ai/trib T2 - port Zj/trib Tb (STS1 link connection #2)
- port Ai/trib T3 - port Zj/trib Tc (STS1 link connection #3)
- port Ai/trib T4 - port Zj/trib Td (STS1 link connection #4)
- port Ai/trib T5 - port Zj/trib Te (STS1 link connection #5)
As both ends of the link build up this same relation information, it is
possible
to use as link connection identifier only one of the two SNPs. E.g. node A
identifies the STS1 link connection #1 by means of "Ai/T1" and sends this
info
to node Z. Node Z receiving this Ai/T1 converts this locally into "Zj/Ta".
Vice
versa, node Z will refer to STS1 link connection #1 as "Zj/Ta" and when
received
by node A, this will be converted into Ai/T1.
For GMPLS this implies that the GMPLS label doesn't identify the timeslot at
both ends of the link connection, only the timeslot at the sending end. The
receiving end needs to convert this then into its local timeslot number.
Regards,
Maarten
Yanhe Fan wrote:
>
> Hi Zhi,
>
> What B does is breaking the label association between A and C by crossing
> connect
> the time slots.
>
> If the cross connect performed by B is fixed (in your case), what needs to
> do is
> just to rebuild the label association by configuration or a protocol such
as
> LMP;
> If the cross connect is not fixed but dynamic, the best thing to do is to
> make B
> a GMPLS speaker.
Tell this to the 1.000.000+ SDH/SONET network elements in today's transport
network... no chance... those will stay non-switched...
>
> Yanhe
>
> -----Original Message-----
> From: Zhi-Wei Lin [mailto:zwlin@lucent.com]
> Sent: Friday, October 19, 2001 6:01 PM
> To: Yanhe Fan
> Cc: John Drake; Kireeti Kompella; ccamp@ops.ietf.org
> Subject: Re: Moving right along ...
>
> Hi Yanhe,
>
> Maybe there is some misunderstanding. Node B is strictly not GMPLS. It
> is simply part of the network that made up the "dumb" pipe. And because
> the control plane control channel is separate, A and C can be adjacent
> without ever knowing who B is or that even B existed. From A and C's
> point of view, the network looks like:
>
> A ----- C
>
> Think of the case of how a BGP sees an area. If an area is made up of:
>
> A -- B -- C
>
> and B is internal router, does BGP "see" B?
>
> Zhi
>
> Yanhe Fan wrote:
>
> > Hi, Zhi
> >
> >
> >>For the label value for SONET/SDH or the label description in the
> >>generic draft, the label is defined as directly representing the
> >>"timeslot". Now if I have a control plane controlled transport network
> >>that goes through a ADM that is not control plane controlled, as:
> >>
> >>
> >>+--------+ +--------+ +--------+
> >>| | ---1--------- | | ----1-------- | |
> >>| | ---2--------- | | ----2-------- | |
> >>| A | ---3--------- | B | ----3-------- | C |
> >>| | ---4--------- | | ----4-------- | |
> >>| | ---5--------- | | ----5-------- | |
> >>+--------+ +--------+ +--------+
> >>
> >>Now imagine A and C is part of GMPLS, but B is just a legacy ADM. So at
> >>A you have labels 1,2,3,4,5 (imagine these as in the SUKLM format) and C
> >>has labels 1,2,3,4,5 incoming.
> >>
> >>Now imagine that at B, it actually has a fixed cross-connect that
> >>connects A-1 (A's timeslot #1) to C-2 (C's incoming timeslot #2), A-2 to
> >>C-3, A-3 to C-4, A-4 to C-5, A-5 to C-1.
> >>
> >>Now when A GMPLS sends a message to C, the label it uses is "1".
> >>However, the connection/LSP associated with timeslot "1" is actually the
> >>label of timeslot "2" at C. </z>
> >>
> >
> > In this case, you're trying to use GMPLS to setup an LSP cross a node
(B)
> > which can't understand GMPLS.
> >
> > Now imagine A, B and C are all OXCs, can you setup an LSP A-B-C by using
> > GMPLS if B can't understand GMPLS?
> >
> > Now imagine A and B are LSRs, and B is a conventional router (can't
> > understand
> > MPLS), can you setup an LSP A-B-C?
> >
> > Now imagine A, B and C are ATM switches, can you setup a svc if B can't
> > understand PNNI at all?
> >
> > My point is that in order to make you case work, some additional work
need
> > to
> > be done besie simply using GMPLS as singnaling protocol, becase some
node
> on
> > the path is not GMPLS speakers. A and C needs the info of time slot
> aligment
> >
> > between them. This can be done by some configurations, some link
> management
> > protocls (LMP maybe) etc. As long as A and C have this piece of info,
> > eveything
> > works fine.
> >
> > Yanhe
> >