Pitfalls of _accepting_ /24s

Fri Oct 17 01:01:11 UTC 2003

A proposal was made some years ago, which I thought was by Tony Li, 
but, IIRC, he says it wasn't original with him.  It does require 
cooperation from competitors, but can reduce the number of 
announcements. Under some circumstances, it may cause blackholing, 
but so can /24 filtering.

The idea is to establish bilateral blocks of provider space. Let us 
say Provider A and Provider B recognize that they have a significant 
number of common multihomed customers.  Arbitrarily, one of the 
providers (assume A) starts off with a block -- let's say a /19 or 
/20 to which both providers will assign their multihomed customers. A 
and B peer and send more-specifics to each other.

To the outside world, however, A advertises its largest aggregate 
plus the multihomed block.  B advertises this block of Provider A 
space as well as its own aggregates.

If A and B do not peer, the likelihood of blackholes become much 
higher since they may not see the more-specifics in the multihomed 
block.

Has anyone reexamined this proposal lately?