> > >I got an even more fundamental question - why does MPLS require IP at
>all?
>> >At the risk of starting a religious way, it's not called Internet
>Protocol
>> >Label Switching, it's Multi-protocol label switching. MPLS has
>effectively
>> >become a feature of IP, as opposed to a generalized control-plane
>mechanism
>> >for which is what it was originally intended.
>> >
>>
>> Let me offer a different way to look at it. MPLS really isn't
>> monolithic. As a sub-IP protocol in the IETF, basic MPLS still has
>> separable forwarding and control plane aspects. The control plane
>> involves path setup protocols such as RSVP-TE and LDP. These, in
>> turn, have to get overall topology information from _somewhere_.
>> Besides IP routing protocols and PNNI, what is there for that purpose
>> that wouldn't need to be invented?
>
>You just hit it on the head. First of all, why is it considered a sub-IP
>protocol? In fact, why is the IETF running the show in the first place?
Because it can, and does.
I've been involved in Formal International Standards Bodies, where
the Camel was developed as a functional specification for a Mouse.
The market and the world are far faster than the carriers would like
it to be.
When I worked for a primarily carrier-oriented vendor, there were
deep emotions that they could make IP go away with:
(1) Ubiquitous fiber
(2) Apparently manually provisioned MPLS, since they equated the topology
to something of equal complexity and hierarchy to what you can do in
SS#7.
>MPLS has potentially far more applicability than just in the Internet (for
>those who didn't catch it, the 'I' in IETF stands for Internet). For
>example, MPLS has tremendous potential for all the world's carrier's ATM
>networks. But right now, for them to take advantage, they have to upgrade
>their ATM switches to IP, rather than just installing a MPLS multi-service
>switch as a dropin replacement.
>
>>
>> Generalized MPLS (GMPLS) is certainly not IP only, as packet
>> forwarding is only one of its modes. It can set up forwarding based
>> on wavelengths, time slots, or ports.
>
>Neither is draft-martini, draft kompella, draft-fischer, or any of the other
>drafts.
>
>But the point is not the forwarding plane, it's the control plane, which
>still relies on IP.
What do you propose as a scalable alternative that doesn't simply
meet telephony needs?
>
>>
>> The first MPLS predecessor, Ipsilon's (now part of Nokia) IP
>> switching was planned as a faster means of lookup than conventional
>> routing. With advances in L3 hardware and software, that simply
>> didn't turn out to be useful or even scalable.
>>
>> Those initial implementations, by Ipsilon, were ATM dependent both
>> for path setup and transport.
>
>And I think this functionality was sadly lost. Not the transport
>functionality, but the path-setup functionality. I think more work needs to
>be done on the ATM side of things to make MPLS more palatable to carriers
>who run lots of ATM and would like to migrate to MPLS but want a smooth
>transition path.
Or some carriers may be displaced by VoX. I've seen quite a number of
marketing research documents that suggest the typical telco wants 90%
L2, 10% L3, because that's what they think their provisioning people
can understand.
The models of manual provisioning, settlements, central coordinating
authorities, etc., still persists in the carrier view of the world.
Also, there are a fair number of vendors that want to retrofit full
MPLS into the spaghetti code of their ATM switches. I've tried to do
that. It was a nightmare. PNNI isn't enough.
Message Posted at:
http://www.groupstudy.com/form/read.php?f=7&i=54582&t=54507
--------------------------------------------------
FAQ, list archives, and subscription info: http://www.groupstudy.com/list/cisco.html
Report misconduct and Nondisclosure violations to [EMAIL PROTECTED]
