[e2e] traffic engineering considered harmful

[Geoff Huston]

At 6/13/01 10:19 AM, Henning G. Schulzrinne wrote:
>grenville armitage wrote:
> >
> > Manish Karir wrote:
> >         [..]
> > > think about it...if you spend millions building
> > > out your network, would you let somebody else control it??
> >
> > If one was in a monopoly situation, the answer is obvious.
> > But perhaps Jon's suggestion makes more sense if you think
> > of the user selecting on a per-AS basis, rather than literal
> > hops inside any particular provider's million dollar network.
> >
> > Start with something simple like an end user architecture that
> > supports minute by minute choice/re-choice of one's first/last-hop
> > provider.  The network operator(s) would get to keep control of
> > *their* networks, but would have to compete to be chosen as
> > *the* network by educated consumers. Scaling the "informed choice"
> > aspect of this model is left as an exercise for the reader...
> > (in the US phone system "informed choice" appears to involve
> > advertising imploring us to "dial 1800CALLATT" or "1010<blah>"
> > for some price point or other. There's a lesson in there
> > somewhere...)
>
>Isn't that what multihoming does, with the edge router rather than the
>user making the choice? How does this differ from (particularly) IPv6
>loose source routing?


well, looking at the as path changes coming in from the edge in the current 
routing table one can already surmise that there is a fair deal of traffic 
engineering-via-routing happening already. Don't forget that out at the 
edge the inbound traffic dominates the outbound load, so traffic 
engineering is often all about biasing the egress path choices made by 
remote AS's towards the local AS, rather than making egress choices for 
traffic leaving the local AS.


   Geoff