On Tue, 27 Apr 2010, João Damas wrote:
the idea here is twofold: 1) to take routing policy out of addressing policy. That part was done
This strikes me as a bad idea. If you can't aggregate a set of addresses into a single announcement, then the best way to keep the number of prefixes in the global tables down is to have people get another block. That way you can have a strict filtering policy, and noone tries to get away with deaggregation because: i) It won't work ii) It isn't needed This is a little extra work for the RIRs - each new non aggregatable block needs another resource request, but it means that we avoid the mess that we've already got in IPv4 where things tend to work as /24s anywhere, so why bother aggregating beyond that. In other words, if an organisation can't announce its address space in a single aggregatable block, then give it multiple blocks - a new block for each disconnected network, rather than force it to deaggregate the block and announce it in little pieces (and thus opening the door for everyone to deaggregate their blocks unnecessarily).
2) make people aware that while de-aggregation is possible, it has its costs, for everyone, not just the de-aggregator, and therefore one should think a bit before splitting an allocation down into atoms.
The problem is not that organisation don't know, it's that they don't care, and there's little reason for them to care! If someone were to deaggregate a /32 into 256 /40s, it doesn't cost them - it costs everyone else. It costs in router FIBs, RAM, and convergence time. Does their deaggreation harm themselves specifically? As long as it seems to work, no. Making people aware it has costs (for everyone else) doesn't stop it happening. Having a recommendation paper isn't going to be enough - it didn't work for IPv4, there's no reason to expect it will work for IPv6. It's also solvable fairly easily as described above. What we do now (potentially opening the door for orders of magnitude of deaggregation) could have rather large consequences in future - even if RAM, TCAM, FIB, etc are orders of magnitude larger in future, convergence time is still increased proportionally (or worse?). Regards James