- Do we want a maximum allocation size ?
Yes, based on the applicant's run-rate. In other words the maximum is not an arbitrary fixed amount of IP space, but an arbitrary fixed time period that is used, with the run-rate, to calculate the allocation size. The end result is that everybody's last allocation runs out at roughly the same time.
If we have typical PA allocations of /26, which by and large will be unroutable, we make the last /8 broadly useless. If we do that, we effectively make the v4 dry period start one /8 earlier.
Any policy change is going to make the v4 scarcity period start earlier so the best we can do is to make the shape of that scarcity period less harsh. This is why I keep talking about an attempt to make everyone's free space run out at roughly the same time. I believe that this will mimimize the negative impact even though it would result in a certain amount of routing table consumption for very long prefixes.
Does this matter ? Well, if our work makes the last /8 useless, it moves the sky a thousand feet closer to the earth. If we don't do this work, then the sky is STILL FALLING.
The sky is falling regardless of our actions. And if RIPE hands out /26 blocks that end up being unroutable, the unroutability is the ISP's fault, not RIPE's. In other words, if LIRs vote for a policy change that results in /26 allocations, one would expect that those LIRs would follow through by changing their routing filters.
- Do I think the minimum allocation size should fall ?
Maybe, but if everyone decides to be a 'good citizen' and ask for a / 22 rather than /21, how much extra time would we buy ? 3 or 4 months ?
Perhaps a smaller minimum allocation size for the last /8 pushes the sky back a few hundred feet, but it sure looks like it's getting closer..
If there is a good reason to simply reduce allocation sizes then there is no good reason why we should wait until the final /8 to make that reduction.
- Do I think companies should be limited to one PA allocation from the last /8 ?
Probably not, because we will make a lot of business for lawyers trying to define what 'one company' is. If a company has an arm in Switzerland, one in the UK, one in France, then does spending money on three LIR memberships earn 3x PA ?
All modern multinational companies, and many larger domestic companies are actually formed as a collection of many corporations. Sometimes this is simply a result of the way mergers and acquisitions are done, and sometimes it is done for tax reasons. These component corporate entities do not necessarily use the mother company's brand name in their corporate name, so RIPE may not even know that two LIRs are related. I think that we will do better by focusing on the demand, and make rules for applicants who ask for address space. Separately, RIPE might want to work on the issues of merging LIRs.
We can do much harm whilst trying to do good in the construction of intervention policies. The only fix is to deploy v6, so policies which are designed to extend the life of v4 will muddy the pro-v6 message that we want to send to operators.
It is still reasonable to change the IPv4 policy so that everybody suffers equally as we use up the last drops of IPv4. --Michael Dillon