On Thu, Jul 17, 2008 at 6:16 AM, Iljitsch van Beijnum <[EMAIL PROTECTED]> wrote: >> I personally think there's still a little room for improvement in the >> FIB calculation. It might be possible to reduce the FIB by selecting >> the most-common exit in the RIB for each /8 and then only installing >> the more specifics in the FIB for the prefixes that use a different >> exit. But the closer you get to the core, the less improvement that >> offers, something you should be able to confirm or refute with RIB >> snapshots from various operators if you care to try it. > > That's because today, there is little rhyme or reason to the distribution of > prefixes within each of those /8s. However, if you know that everything in > 145.0.0.0/8 is in the Netherlands, there is really no reason to keep all > those more specifics in your routers in Seattle, as interconnection with > Dutch networks typically happens in Amsterdam with backups in England and > Germany, or at the very least in the US north east, where all the > transatlantic cables land.
Iljitsch, Sounds like a straightforward research project. Go get RIB snapshots from core routers at various tier-1's. Compute the maximum aggregation you could get in the RIB to FIB computation with the prefixes as they are now. Then match whois records to the prefixes and reallocate them by a manageable geography, "country" let's say. Simulate the incremental nature of address assignment with a requirement that the prefixes within a country be randomly distributed within that country's block, taking no hints from general topography. Then remap the reallocated prefixes back to the originals in those RIB snapshots and see what aggregation you achieve in the FIB to RIB computation. I look forward to reading your results. *Note* that while you can potentially drop the routes in the FIB, you can't drop them in the RIB. Your downstream customer could reasonably have an alternate set of routes where some of your more specifics are optimal and some aren't. Even if two adjacent routes in your router are identical in every respect, his may not be. If you drop the routes from the RIB, it impairs his ability to make good routing decisions. Since the FIB costs *a lot* more than the RIB, reducing just the FIB is still an interesting endeavor. Regards, Bill Herrin -- William D. Herrin ................ [EMAIL PROTECTED] [EMAIL PROTECTED] 3005 Crane Dr. ...................... Web: <http://bill.herrin.us/> Falls Church, VA 22042-3004 -- to unsubscribe send a message to [EMAIL PROTECTED] with the word 'unsubscribe' in a single line as the message text body. archive: <http://psg.com/lists/rrg/> & ftp://psg.com/pub/lists/rrg
