RE: What frame relay switch is causing MCI/Worldcom such grief?

Martin Cooper said something similar to:
> But the reason ISPs adopted NBMA L2 networks in the first place is
> because the telcos convinced them that L2/L3 integration was going
> to happen.

Actually, that turns out not to be the case. The adoption of layer 2
switching technology into the internet core set of tools predates any of
the ATM hype. It was done with frame relay.

Oh absolutely - I wasn't just referring to ATM in the context of NBMA
L2 switching.

It was not done on the
basis of L2/L3 integration, but because it gave the ISPs something more
useful than just a fixed size (DS3) set of pipes between places.

Well ok, but if you follow that marketing pitch (making nailed up
L2 links more useful to extremely popular overlayed L3 networks
like the Internet) to its logical conclusion, it ends up there,
because the motivation behind it is to be seen to add value to
the product (shifting unspecified data from A to B) to stave off


Sorry - NBMA L2 networks are really L3 networks. There is not theoretical
difference between FR network with OSPF routing and IP CORE network with
OSPF routing. Fixed-sized ATM cells make switching easier but I do not
think it's of any importance for the modern hardware developers, and the
announcement CISCO did about developing AIC's for the CEF (L3) switching
just prove it.

You are talking abouth the wrong things. The real MLPS and L2/L3 idea was
to divide core and customer's routing into 2 independent levels - core
routing should work with N nodes (where N is the number if your
routers/switches) and should be mono-vemdor, mono-protocol system, and
the customer's routing should work with the unstable, multi-vendor and
multi-protocol Internet (and private networking) world. Just this is the
real reason for this L2/L3 playing.

MLPS just realise this idea and mixture this idea with _classify and mark
QoS parameters_ idea - just both great ideas. But when people add AMT QoS
there, with a lot of _pvc bundles, svs's , etc etc_ they fraud themself -
it's possible to build this, but it's not easiest way to do it. For
example, if you built some kind of 2 level IP-like (circuit-less)
network, and translate QoS parameters into some kind of IP TOS and
and avoid complex and huge BGP tables inside the core network - you can
achieve QoS just by existing tools (preferences, RSVP if you want it,
WFQ, etc etc) and you can avoid useless QoS calculations over the huge
network while you have not congested links in it (if you use queue-based
QoS control it work only when it's nessesary, if you use circuit-based
QoS you should calculate QoS everywhere and forever even if you have not
congestions at all). But it's offtopic for this thread, sorry - a lot of
people are tired of this discussions and ATM/MLPS/QoS words... The most
important idea in L2/L3 division some theoretics are doing now is to
divide CORE/CUSTOMRE's switching into the 2 independent levels, and use
the simplicity of the core backbones (plain, single-vendor usially) to
make the packet switching more stable and effective.