[arin-ppml] Draft Policy ARIN-2023-2: /26 initial IPv4 allocation for IXPs

[Bill Woodcock]

> On Jun 21, 2023, at 10:18 AM, Matt Peterson <matt at peterson.org> wrote:
> It's clear this proposal did not receive feedback from those of us who operate IXP's (or those who lived through the ep.net era). Renumbering events are often multi-year efforts for an IXP, this "savings" is not worth the operational overhead. I'm not in support of this proposal. This is a solution looking for a problem, we have both the appropriate pool size and a method to refill.
> If anything, the 4.4 requirement language around "other participants (minimum of three total)" could use some attention. ARIN's service region has many "shadow IXP's", which may have 3 unique ASN's (say a route server, route collector, and management network) - but are all operated by the same organization. That does not seem like a legitimate definition of an exchange point, especially when that operator is the only participant over several years.

I would just chime in to say that I definitely agree with the first of Matt’s points.  IXPs rarely know how long it will be until they need a larger address space, and when they do, it’s typically too late to renumber hundreds of different organizations.  This policy would create a vast amount of unnecessary work for ARIN members, while yielding no obvious benefit to anyone.

As regards what constitutes a “real” IXP, that’s a tougher question.  While Matt is undoubtedly correct that Andrew’s Basement Exchange, the canonical example, may not have three unrelated participants at the time they apply, that’s also true, at the outset, of many IXPs which follow a sure-footed path to success.  And it’s not possible to know in advance, in any sort of replicable policy way, which will ultimately succeed, and which fail.  I guess my thought on this is to be liberal in distributing, and also relatively quick to reclaim when an IXP goes defunct.  Over the last 31 years, 34% of all IXPs that were established, went defunct.  Many took years or decades to fail, but once they have, there’s no reason for those allocations to persist.  Most IXPs are one-off, a group of ISPs who get together to form an IXP in a specific location.  A few consist of a single organization spanning multiple unconnected locations…  When those organizations are for-profit startups, I guess I would evaluate their claims carefully, and do a slow-start, rather than allowing them to use the policy to get a /24 for every new claimed location, in advance of proving themselves.

                                -Bill