Hi Noel,
> In prior conversations with you, you have suggested that the broad > architectural approach I prefer for path selection - i.e. topology > distribution, with unified path computation (although perhaps 'monolithic' > is a better term than 'unified', since the latter term was used elsewhere > in routing) - is problematic. I don't recall your reasoning exactly (not > that I ever got it in detail, I don't think), IIRC it was something to do > with a combination of: > > i) ISPs didn't want to hand out topology information And perhaps more importantly, they are unwilling to disclose their policy information. Peering policies are a very touchy subject. Being nice to competitor A and not nice to competitor B is the type of diplomatic fodder that starts corporate battles. > ii) ISPs like the kind of policies about traffic flow they can impose > with destination vector architectures Specifically, the ability to do hot-potato and cold-potato routing, as well as destination originated traffic engineering. > iii) the overhead of path computation needs to be distributed This is the least of the issues. Yes, this needs to be distributed for scalability. But then I come from a church where _everything_ needs to be distributed for scalability. > So perhaps a discussion about what you see as the shortcomings of that > approach, and discussion about whether those issues are or are not > handlable, would be a good thing to do here? So I'm more than happy to have that conversation, but I would very much like it to proceed into a concrete topic proposal if we're doing it on RRG. If we're having just a fun architectural discussion, then routing-discussion might be more appropriate, so that this list could be reserved for those developing topics. Warmest regards, Tony _______________________________________________ rrg mailing list rrg@irtf.org http://www.irtf.org/mailman/listinfo/rrg
