[address-policy-wg] 2023-01 New Policy Proposal (Reducing IXP IPv4 assignment default size to a /26)

* Matthias Wichtlhuber

> regarding 6.1(4) - (default is /26, /25 is handed out upon request):
> 
> The idea behind this part of the proposal is that there are two kinds
> of IXPs, one that is founded due to the need to interconnect a few
> networks in a specific region and no intent of growth and one that is
> operated with the explicit intent to grow and generate business. IXPs
> can decide on their plans on their own and do an appropriate
> selection of their initial assignment.

That places a big trust in the new IXPs to not ask for a /25 if they do
not need it. To me, that seems naive - there are strong incentives for
new IXPs to ask for as much as possible regardless of actual need. You
point them out in the proposal yourself: avoid and/or postponing future
renumbering and IPv4 purchases.

A similar situation exist in IPv6. LIRs can freely choose initial
allocations in the range /29../32. I bet the vast majority of new LIRs
do not need anything larger than /32 (they definitively do not have to
worry about renumbering or IPv6 purchases), yet in 2022 the numbers
were:

1035 /29 allocations
3 /30 allocations
0 /31 allocations
64 /32 allocations

So 94% chose the biggest they could get. That's human nature for you.

Under the proposed policy I bet you will see the same thing happening,
most IXPs will ask for /25s. (I sure would, if I was to start an IXP.
There is simply no reason not to.)

> regarding 6.1 (7) - (assignments down to /27):
> 
> I agree that the policy can be gamed by asking for a /27 and jumping
> to a /24; we might consider to restrict this part to only handing out
> /27s once the IXP pool is depleted, but not upon request. I doubt
> RIPE has handed out any /27 assignments upon request anyways; maybe
> someone from RIPE can look into this?

For what it is worth there are currently 32 /27 assignments in the
delegated file (52 assignments /27 and smaller), but none of them are
from the IXP pool. (They could still be used by IXPs though.)

In fact, there are exactly 0 assignments smaller than /24 made from the
IXP pool, even though the policy allows for this. This makes me even
more convinced that IXPs will generally chose to take the largest
prefix policy allows them to.

> regarding the general point of lower bound of assignments and your
> comment:
> 
> > Alternatively we would need to consider the supporting data for
> > this proposal bogus.
> 
> As I have stated a couple of times in the previous discussions, the
> data in PeeringDB can only provide a lower bound of connections per
> IXP; nevertheless it remains the best dataset we currently have. Thus
> the decision will have to be a mix of listening to experienced people
> from the IXP community and the results of the analysis. I think a /26
> is a good compromise here.

I was not trying to restart that part of the discussion. We can agree
to disagree on whether or not /26 is a good value, and leave it at
that. (It's better than /24, we can agree on that.)

However, this comment was made in the context of my proposal of adding
the «IPv4 space dust» to the IXP pool, for use when the pool runs out
of /26s.

The NCC inventory currently contains 47+ blocks in the range /29../25
that *would* have been perfectly usable for assignment to IXPs, but
they can not be assigned to IXPs under this policy because they are not
part of the specially designated /15 IXP pool. I say let's add them, so
they actually can do some good, instead of rotting away in the NCC
inventory.

This «IPv4 space dust» also includes some prefixes smaller than /27,
which would not be assignable under the proposed policy even if had
added the «IPv4 space dust» to the IXP pool. However, there *are* small
IXPs that could potentially have used these. There are multiple
examples just here in Norway:

Trondheim IX: 5 members
Bergen IX: 6 members
Tromsø IX: 3 members
Stavanger IX: 8 members

Source: https://www.nix.no/who-is-connected/ (bottom of page)

All of them would make do with a /28, and three out of four would make
do with a /29. I would further note that these are not newly started
IXPs experiencing rapid growth, they are well established.

In my opinion, denying a small IXP similar to these the opportunity to
receive a small assignment – especially when such small prefixes are
the only ones left in the NCC's inventory – just because some folks
from DE-CIX do not «consider [it] to be an IXP» just seems cruel. I do
not see any harm whatsoever in allowing their assignment.

Tore