On Wed, 2005-08-17 at 13:49 -0400, Jeffrey Haas wrote: > An example of this corner case is where the same AS has more than one > instance of an anycast service. In one vendor's BGP multipath > algorithm, if you select more than one path to this destination in > the same AS, you may still have traffic that may reach different anycast > node instances.
Jeff, If the originating AS advertises a single mapping (prefix -> server) to its neighbors, the downstream ASes will not override this decision. It is up to the originator to control where the traffic gets received. As for the issue of consistently selecting one of multiple paths... take one unicast routing advertisement from AS X. If there is multipath downstream and a consistent path is not selected for a given 5-tuple (src,dst,proto,port#s), plain IP features like PATH MTU discovery may not work correctly. The problem exists independently of "anycast". > > While this is indeed a corner case, it relies on that particular vendor's > implementation to "get it right". Given that this is a non-standardized > feature All depends on one's definition of standard. I'd argue that BGP multipath "works consistently among different vendors", despite the lack of an IETF approval stamp. > and that I've heard several non-optimal multipath suggestions, > this document would be well served by having some sort of caveat in it. > > E.g.: > > "In cases where a BGP implementation supports multi-path exit selection > in such a fashion that it is possible that routes to different Anycast > Nodes may result, that implementation SHOULD support a feature to > disallow multi-path route selection for Anycast reachability." There is no reason why BGP needs or should treat "anycast" different from unicast... in fact it MUST NOT know whether there is more than one advertisement for a prefix. This kind of text is harmful. > > Effectively, I think this means "the bgp routes should be almost identical > and that hash-based ECMP traffic forwarding" are requirements. > Putting such a thing in this document is probably inappropriate. :-) Pedro. _________________________________________________________________ web user interface: http://darkwing.uoregon.edu/~llynch/grow.html web archive: http://darkwing.uoregon.edu/~llynch/grow/
