[ixpmanager] route server template/32bit ASNs

[Nick Hilliard]

On 06/08/2015 14:13, Stefan Kaltenbrunner wrote:
> On 08/06/2015 03:03 PM, Nick Hilliard wrote:
>> 1. static mapping doesn't scale
> 
> I agree there but for our usecase we probably dont need that level of
> scaling for the forseeable future...


>> 3. INEX did some models based on this a couple of years ago and decided
>> that it wouldn't work
> 
> could you elaborate on "wouldn't work"? if it works for AMS-IX (disclaimer:
> I only read their docs so far and talked to some active users of the
> feature there) - it should work for most IXPs...

Check out pages 31-32 of:

http://ripe59.ripe.net/presentations/hilliard-irrtoolset.pdf

These two tables show the difference between published actual BGP sessions
and peering relationships as described in IRRDB policies.  You can see that
there is virtually no relationship between the two.

There would be a lot of work to get table #2 in the same shape as #1.

>> 2. is potentially interesting but is limited by the fact that not all
>> routers support draft-ietf-idr-as4octet-extcomm-generic-subtype, and many
>> only support a small number of fixed types of extended community.  This
>> means that interoperability is a problem.
> 
> hmm - I dont assume there is some sort of list available on who does and
> who does not?

AFAIK, only de-cix supports draft-ietf-idr-as4octet-extcomm-generic-subtype
on their route servers.

>> There are no good solutions to this problem at the moment.  INEX advises
>> route server users to use ASN16s, but that can mean renumbering ASNs.
> 
> yeah - we are talking a small IXP starting up here that has a large number
> of "first IXP we are connected" participants with ASN32 on one side (and
> having to renumber would be way too much work for them) and a number of
> fairly large ISPs that need that kind of control because of downstream
> agreements with some of the participants.

We have similar issues.  Agreed that renumbering ASNs is a pain, but ISPs
who use ASN32s should bear in mind that they will also be unable to use
exportable BGP communities on their networks.  This will cause them scaling
problems in future.

Nick