Hi, On Fri, May 03, 2002 at 08:45:55AM -0700, Michel Py wrote:

...

Maybe. But I'm not so sure whether this is so terrible - if you do not want to see /48s, filter them out, and live with potentially non-perfect routing (to the aggregate and find the more-specific there).

I'm sorry, but what do you achieve in terms of multihoming doing this? If you filter PA specific routes, the only aggregate you see in the DFZ is that provider's block, effectively making the site singlehomed.

Among those that do not filter you, you have ideal and perfect connectivity. If you are filtered wherever and for whatever reason (or flap-dampened or whatever), you are *still* reachable over the aggregate, maybe not as redundant and direct as otherwise, but as a good upstream usually has "many 9's" today, this should be sufficient. If you cease to be multihoming (for whatever reason), the route disappears from the global table again. This is why this is a recommended (and frequently done) for doing IPv4 BGP multihoming today (get a block out of your upstream's PA, instead of using PI), and among the alternatives available *today*, I consider this to be the least evil. If I've overlooked a flaw, and there is a better way *today* that does not instantly create non-removeable legacy problems, please point this out without resorting to "what you are doing is evil, a remedy will be available at some uncertain point in the future" (because while that may be the ideal academic solution, some people actually want to use IPv6 *today*). Gert Doering -- NetMaster -- Total number of prefixes smaller than registry allocations: 45077 (47584) SpaceNet AG Mail: netmaster@Space.Net Joseph-Dollinger-Bogen 14 Tel : +49-89-32356-0 80807 Muenchen Fax : +49-89-32356-299