A couple of additional comments which I should have added in the first place... Iljitsch van Beijnum wrote:
On 23-mei-2007, at 11:47, Wilfried Woeber, UniVie/ACOnet wrote:
http://www.ripe.net/ripe/meetings/ripe-54/presentations/ RIPE_NCC_Statistics.pdf
The bottom line is that the # of PI assignments has (considerably) surpassed the number of PA assignments since 2003, and that the load on the routing table for PI is thus bigger than for PA, although the *percentage* of PI space as compared to PA is approx. 2%.
(As a percentage of the address space, not a percentage of the number of blocks.)
Correct.
Or, the other way 'round, we use more than 50% of (additional) routing table slots for some 2% of address space (PI) and less than 50% for some 98% of PA.
Which of course is not necessarily the full story as there probably are filters in place to prevent a good number of them to show up in the DFZ.
And that's with IPv4, where you have to show you really need a block of 256 addresses to qualify for a PI block.
Minor correction: you don't have to require a /24 equivalent to get PI. Actually, that is the (imho important) cross-link to the proposal for "upgrading" any smaller PI assignment to a /24 "if there are routing problems". Ref: http://www.ripe.net/ripe/policies/proposals/2006-05.html Otoh, going classful again for PIv4 would change the 2%/98% ratio ;-)
In IPv6, that hurdle has been removed so it has the potential to see even larger numbers of PI as soon as IPv6 deployment starts taking off.
Wilfried.