In message <[EMAIL PROTECTED]> Iljitsch van Beijnum writes: > > On 21-mrt-2007, at 14:45, Curtis Villamizar wrote: > > >> Does this address the solution where multihomer M uses ISPs A and B, > >> and M's prefix is injected into BGP by both A and B and NOT by M? > >> I.e., "inconsistent origin AS", which is frowned upon. > > >> If not, the text is unclear. If so, why is there no discussion of the > >> normal situation where the multihomed AS advertises its prefix > >> itself? > > > If normal means "architecturally pretty" then the multihomed prefix > > advertised by the true oridinator is normal. No special case is > > needed for that. > > If you mention special cases you must also specify the normal case. > These are engineering documents, any time someone has to think about > semantics we're in trouble.
OK. I see your point. IMHO Adding mention of the normal case would be a better fix than removing this text. The remainder of this thread is an aside to the discussion of whats in the draft and so can be disregarded. > > Many multihomed enterprises don't have their own AS and have a single > > prefix. They don't run BGP. Instead each provider conditionally > > originates the prefix on their end with their AS if the link is up. > > If normal means "more common" then this might be the normal case. > > Absolutely not. Last time I checked the first figure for incosistent > origin AS were less than 1000 prefixes. The nature of this practice > means you'll never see every affected prefix from any one given > vantage point but it's pretty obvious that this is NOT a common > practice. Good. Seems less common than it used to be. > And why sghould it be? An AS number is much easier to get than > address space and running BGP is much simpler than alternative > methods to inject/remove prefixes depending on connectivity to a > particular ISP. I suppose my observation of this trend is severely dated and no longer valid. Dating myself again. Enterprises used to have an easier time getting PI space and they didn't trust the Internet so some wanted multiple providers. Though there was no evidence that multiple providers yielded higher availability that two circuits to the same provider. In fact the opposite was true since the provider bought the circuits and with same provider they could insist on diverse paths (and not always get it, but always get charged for it). Curtis _______________________________________________ Sidr mailing list [email protected] https://www1.ietf.org/mailman/listinfo/sidr
