On Tue, 16 Aug 2005, Pedro Roque Marques wrote: > > I believe that the above part of Dean's criticism is valid, and I > > think that it would be worthwhile for a section on per-packet load > > balancing to be added to the draft. > > I'd argue that it is not. That in fact load-balancing is not > in play when anycast is used. Inter-domain routing (bgp) will > deterministically pick one of the available paths. > > I may be wrong. But i've yet to see any explanation of how multiple > paths would be acceptable for load-balancing.
The people who want to do per-packet load balancing across multiple paths might not be using BGP in the way you expect. For example, they might have multiple default routes to multiple upstream providers, and they might try to load balance over the default routes, using some mechanism outside BGP. They might just have static default routes to their upstream providers. They might be using a private extension to BGP that allows multiple routes to the same destination. They might be in the future, using some routing protocol not yet invented. If packets addressed to the same destination IP address are load-balanced across multiple default routes that end up going to different neighbouring ASes, and if the destination IP address is that of an anycast service, then different packets can end up going to different instances of the anycast service. This will not work. So people should be advised not to do it. --apb (Alan Barrett) _________________________________________________________________ web user interface: http://darkwing.uoregon.edu/~llynch/grow.html web archive: http://darkwing.uoregon.edu/~llynch/grow/
