On 04/02/2019 15:02, Sander Steffann wrote:
> Hi Daniel,
>
>> But how tenable is it both in principle and in 'Internet governance'
>> terms for the NCC to collect fragmentlets of IPv4 and just sit on them?
>>
>> Not very!
>>
>> So we need a policy to allocate them in a useful manner.
>>
>> The question before us is: What is the minimum useful allocation?
>> Nothing else.
>
> You are much better at summarising than I am :)
>
> Andrea Cima has shown us at RIPE76 that /22 is not useful, and /24 is just
> about useful.
Sorry for not being precise. I meant 'useful to route packets' and not
useful to make the allocation process more convenient.
So let us look at what minimum prefix is useful to route packets:
Looking at http://bgp.potaroo.net/as6447/ it appears that at this time
more than 50% of the IPv4 prefixes seen by that collector are /24s. So
/24s are useful. The numbers of prefixes longer than /25 are negligible
in comparison. Other statistics Geoff provides there also support that
/25 or longer is not useful in practice today. Geoff's data agrees with
what RIS sees too. This should be no shocking news to anybody here.
It is not tenable for the NCC to force new LIRs to wait for a /22 to be
returned when they would be happy to use a /24 that the RIPE NCC has.
So that needs to be an option. Offering the choice to wait for a longer
prefix raises all sorts of complications about fairness and
practicality. Therefore a rational policy will end up at
one-size-fits-all: /24.
It is just the reality of current routing practice. A rational policy
will have to accept it.
Daniel
