Well, if the AS-set doesn't fit, that's a problem. But it isn't necessarily the case that simply because an AS that gets, say, RIPE addresses is transcontental means that it can't be part of a "RIPE" aggregate. One would have to sit down and see how bad the paths are relative to the RIB savings. I don't suppose anyone has done such a study? PF
-----Original Message----- From: Florian Weimer [mailto:fw@deneb.enyo.de] Sent: Monday, September 29, 2008 7:31 PM To: Paul Francis Cc: Routing WG Subject: Re: [routing-wg]a (perhaps) naive question...
* Paul Francis:
I was wondering the other day why we don't do aggregating in BGP along the lines of registry assignments. For instance, what stops the set of ISPs within Europe from from taking the 15 or 20 prefixes given to RIPE by IANA, and collectively aggregating those prefixes when advertising to non-RIPE ISPs? It seems to me that they could advertise all of their AS#'s as a huge AS-set for these prefixes.
The latter probably doesn't work because the AS set size would exceed the maximum size of a BGP UPDATE message. The number of prefixes is also large than 20 due to the pre-RIR swamp space.
Is there a technical issue that prevents this, or does the organizational effort needed simply outweigh the benefit that would accrue?
It leads to worse routing decisions and does not reflect the reality of autonomous systems spanning multiple continents. (Anyone who is multi-homed to one of those trans-continentals would probably need a globally visible prefix.)
Actually, for ISPs in the RIPE region, it would make more sense to aggregate the prefixes of ISPs *not* in the RIPE region.