[c-nsp] mpls-te - dynamic latency?

[Oliver Boehmer (oboehmer)]

christian.macnevin at uk.bnpparibas.com
<mailto:christian.macnevin at uk.bnpparibas.com> wrote on Thursday, April
06, 2006 5:38 PM:

> Sadly, the prefixes aren't going to be that easy to extricate from
> one another. We have no guarantee that a given subnet won't contain
> hosts who want market data / soft phone / video conf  / message block
> data as well as web and email  
> traffic. Hence, our best bet is a routing mechanism which can forward
> based on dscp marking. 

Then you need to build parallel tunnels between your P/PE routers and
use Class-based tunnel selection to steer the traffic accordingly. This
is currently your only choice. Depending on how your network looks like
and where the critical links are, you might not need PE-PE mesh, P-P
tunnels in selected areas might be enough.

> I can't be the only one - there must be loads of people on here who
> can see a need for layer 4 routing in large networks. Most
> organisations want to be able to use all the links they've paid for,
> and this gives them the chance to do it, and remove invisible
> mechanisms like PBR, and potentially sidestep requirements for more
> technology like TE.   

I don't want to argue, this is what MTR addresses, and this is one
reason it's being developed. I am not sure how MTR links with
MPLS-VPN/VRFs, need to check.. 

	oli