[address-policy-wg] 2019-05 New Policy Proposal (Revised IPv4 assignment policy for IXPs)

Aris Lambrianidis wrote on 04/06/2019 18:03:
> Consulting the PCH IXP directory as Nick did earlier (as well as the 
> Euro-IX one), I think it is also reasonable to say that
> allocating a /24 is ambitious for the overwhelming majority of cases.
> 
> Only 64 of listed IXPs have equal to, or more than, 100 participants, 
> out of 958 IXPs, or about 6.6%.
> 
> In this light, perhaps the default allocation discussed in 6.1.4 should 
> go down to a /25.

/25 is too small, even for smaller IXPs.

~400-500 of the entries in the PCH IXP directory are defunct.  For the 
remainder, the participant numbers are inaccurate, mostly on the low 
side. A figure of about 500 active IXPs is largely corroborated by the 
IXP DB (650 entries, with some effectively defunct).

The figure you need to look at is 50% usage rather than 100% usage.  If 
you pin the assignment size to /25, then 50% of /25 is 64 participants, 
i.e. about ~20-25% of IXPs, not 6.6%.

The current run-out rate for the RIPE pool is about 15k addresses per 
day.  This means that a /16 is 4 days worth of allocations at the 
current rate.  A /16 pool gives adequate breathing room for core 
internet infrastructure, with a /24 assignment size.

The central question of this policy update is not the assignment size 
for IXPs, but whether it's worth investing 4 days out of 30 years worth 
of allocations in order to provide important flexibility for the 
internet core in the future.  I'm inclined to think it is.

Nick