tunneling over IP vs. over MPLS [was: Re: [c-nsp] Cisco 7600 Vs Cisco 6500]

[Simon Leinen]

Saku Ytti writes:
> On (2005-07-13 08:45 +0200), Mikael Abrahamsson wrote:
>> Well, you got me there. We don't do L2TP at all, I see tunnels over
>> IP for bulk packet path as inherently flawed way of doing
>> business. Much more expensive than just routing them.

>  I think our industry is funny in this way. Why are IP tunnels
> odd/bad and perform badly while MPLS tunnels are good and perform
> well?

First of all, I fully share your view.  I would much prefer if our
platform supported (Eo)L2TPv3 - then we could just run a simple IP
backbone and encapsulate Ethernet frames in IP packets at the edges
where needed.  Instead we have to run MPLS in the backbone just to
support Ethernet VPNs.

> CSCO support pwe3 and 'MPLS VPN's over L2TPv3, what does this
> exactly mean? This means, that you could finally actaully buy cheap
> and dump core switches, like you used to with F/R and ATM. Do we
> have cheap and dump MPLS switches? IP lookup is commodity and in
> some ways simpler than MPLS, as IP lookup is hiearchial and you
> don't have to swap IP addresses.  So doing MPLS lookup engine in
> dirty cheap TCAM stuff, I don't know.  But even today, you can buy
> the cheapest L3 switch with IP routing ablity, build your el'cheapo
> ghetto 10G core with loop0 routing only in core and run everything
> over L2TPv3, like it was your MPLS, just without labels. No problems
> doing your 'MPLS VPN' between any two IP islands.

I have the impression that newer "el-cheapo" hardware will do IPv4,
IPv6, and MPLS forwarding anyway (witness the Broadcom StrataXGS
(5660x) or Greenfield chipsets).  Rewriting the labels is not such a
big deal.  These new chipsets also seem to perform Ethernet-in-MPLS
en-/decapsulation rather than L2TPv3 :-(

Since all the carriers love circuits in their backbone anyway and have
bought into MPLS long ago, I guess this all makes business sense.
-- 
Simon.