[ppml] Policy Proposal: Decreasing Exponential Rationing of IPv4 IP Addresses

[Scott Leibrand]

Ok.  So to paraphrase, you anticipate that the rationing will lead to 
more stringent review of IPv4 applications, and that once the IPv4 
addresses available for a particular period 
(month/quarter/year/whatever) are used up, applications will be 
waitlisted and filled on a FCFS basis at the start of the next period.  
That seams like a reasonably fair way to do things, at least until the 
waitlist starts getting long, at which point it might be worthwhile to 
start partially filling requests and require applicants to reapply and 
get back in line as soon as they use their partial allocation.  It might 
be worthwhile exploring such options as part of the policy process, but 
I can also see the wisdom in allowing ARIN staff the leeway to manage 
the rationing  as needed based on changing conditions.

One thing I think we ought to consider is adjusting the timeframe.  
(I'll use your "divide by N" approximation, as it's easier to think 
about.)  Under your proposal, with N=10, I would anticipate that we 
would begin seeing the impact of rationing immediately, and that the 
impacts may be large.  Another option would be to set N to a lower 
value, say 5.  Since we're approximately 5 years away from exhaustion at 
current projected rates, that may mean we can "ease into" the rationing 
regime, which may be advantageous.  If we did that, we wouldn't run out 
of IPv4 space in 5 years, because each year the ration decreases 
exponentially, so that the smaller and smaller free pool continues to be 
extended until IPv6 adoption takes off and IPv4 space starts getting 
returned faster than it's given out.

Another consideration is what to do about global coordination.  As it 
stands today, the large pool of free IPv4 space is managed by IANA, 
which gives it out to RIRs as they use it.  If ARIN were to adopt a 
strict rationing regime, say with N=10, but the other RIRs did not do 
so, then ARIN would see most of the remaining free pool used up by other 
RIRs, which ARIN became less and less able to fill its members IPv4 
requests.  Because of this concern, I think it's important to define how 
ARIN's rationing interacts with other RIRs assignment practices. One 
approach would be to try for a globally coordinated rationing system.  
Another would be to push for IANA to divide the free pool up in advance, 
and delay stringent rationing until ARIN has a fixed-size pool to work 
with. Do you have any thoughts on how to manage that aspect of things?

With regards to markets, I agree with you that an unregulated free 
market in IPv4 space would be a disaster.  I foresee something more 
closely managed, where perhaps someone wanting IPv4 space would first go 
to ARIN, get their justification approved, and then would have the 
option of either getting on the waitlist for available space or buying 
space, up to the amount they were able to justify, on a market.  ARIN 
would then approve the sale, update the registry, and take the applicant 
off the waitlist.  In any event, I see an IPv4 market as something that 
will only be useful once we exhaust the IPv4 free pool, or start a 
stringent rationing regime.  So we should probably focus on proposals 
like Soft Landing and Rationing to deal with the runout phase, and then 
start thinking about markets, reclamation, etc. as soon as we have good 
runout policies in place.

As for contested space, I agree that we need to put policies in place 
that constrain any possible unethical behavior, regardless of whether we 
agree or disagree on the ethics of various individuals or their past 
actions.  I think the public policy process has largely done a good job 
creating such policies to date, and have no problem with your contested 
space clause in this policy proposal.

-Scott

Dean Anderson wrote:
> On Mon, 20 Aug 2007, Scott Leibrand wrote:
>
>   
>> Dean,
>>
>> Thank you for starting this discussion with a policy proposal.
>>
>> IMO we'll need to add some additional guidance to a policy proposal like 
>> this in order to ensure it's implemented in a way that meets the 
>> proposal's goals.  For example, here are a few questions we might want 
>> to address:
>>
>>     * How should rationing be achieved?   Should all applicants receive
>>       smaller blocks than their justified usage would otherwise permit? 
>>       Should all applications be placed on a waiting list and filled in
>>       a first-come-first-served basis as soon as the rationing function
>>       allows?
>>     
>
> My perception of the actual inner working of ARIN right now is that
> requests are handled roughly first come, first served, with the
> exception that people that get all their documents in order first, get
> served first.  I can't say that this is really the case.  But to the
> extent it is the case, then we already have a system that will work for
> rationing.  Rationing based on a hard limit isn't always fair, but it is
> often necessary.
>
> For example, I recently saw on the news of water rationing after a flood
> in the UK. The people who get in line first, get their [ration] bottle
> of water first. This continues until the water buffalo [water tank on
> wheels] runs out.  That's unfair to the people at the back of the line.  
> But there is nothing better, sometimes. Likewise, I don't think we can
> determine who is 'most worthy' other than those who get in line earliest
> and get their paperwork in order first.  One ISP is just as worthy as
> the next to get IP Addresses, assuming their documentation is the same.
>
> But there are a number of natural side effects that are quite
> beneficial. To give example of such side effects I'd like to relate an
> article I read just today about captive insurance strategies in Fortune
> Small Business Magazine. Captive insurance is where a group of small
> businesses get together and raise capital for their own insurance upto
> say $500k, and then purchase reinsurance for larger amounts.  This also
> puts them in control of costs of small claims, because they select the
> claims to pay and the claims to fight, and they also get the profits
> when premiums paid in exceeds costs. Employees of these companies also
> naturally get the message that accidents are paid by the company, not by
> some nebulous far-off wealthy insurance company.  Safety improves and
> claims decline. The company benefits, the employees benefit.  These are
> beneficial side effects of captive group insurance.
>
> Imposing any hard-limit rationing regime will also have side effects on
> the way ARIN staff perform their duties even without further policy
> changes.  Inclination to toward giving out somewhat smaller blocks and
> looking more closely at existing allocation and usage are natural ways
> to serve more people with less resources, and can be done within
> existing policy. I expect that this will be a natural consequence of a
> hard limit in any case.  Once the ARIN staff realize they have a hard
> limit, they'll naturally look harder at documentation.  This doesn't
> have to be specified.
>
> And if tightening doesn't 'just happen' as a side effect, the hard limit
> will stop delegations for a little while, and ARIN staff will simply
> have more time to look at the current requests and documentation.
>
> And if that still doesn't work, we can adjust the policy. However, we
> won't be running out of space for 10 years, so we have some little more
> time to work this out.
>
> I would expect that people who put in requests do not lose the place in
> line just because the hard limit for a timeframe is hit.
>
> I also thought some more about Mr. Herrin's assertion of ambiguity this
> afternoon.  Although I suspect Mr. Herrin may merely be uncomfortable
> with calculating with exponentials, there are actually several different
> ways to calculate e^(-x).  One could use their scientific calculator;
> use Maxima, Matlab, Mathematica, etc; use a series expansion; or use a
> table found a book.  Rounding might make some small differnce.  
> However, none of these make any difference to the 'big picture' of
> preventing IP Address Exhaustion for at least 10 years.
>
> In the first example I posted of a decreasing exponential, I described
> rationing a hundred widgets over 10 years. In the first year, you divide
> 100 by 10, and so give out 10 in the first year. 90 remain.  In the
> second year, divide 90 by 10, and so give out 9. And so on. This
> approximation is fine, too.  So, I think the method by which they
> calculate and scale e^(-x) makes no difference to the big picture of
> keeping address space available for at least 10 years.  I think that
> staff will find some convenient means for doing this. And unless there
> is some "approximation" that isn't really e^(-x) and causes ARIN to
> allocate more IP addresses so that we will run out in less than 10
> years, I don't think it necessary to impose much on the method or scale
> factor for calculating e^(-x).
>
> But if there is still ambiguity, I suggest the following should be
> completely clear:
>
> AIP is the available IP pool at year 0 (the start)
>
> Year 0 is (AIP) * (1/10) * e^(-0/10) = AIP/10 * 1
> Year 1 is (AIP) * (1/10) * e^(-1/10) = AIP/10 * 0.9
> Year 2 is (AIP) * (1/10) * e^(-2/10) = AIP/10 * 0.8
> ....   
>
> If you want to go by 3 month periods: (40 3 month periods in 10 years)
>
> quarter 0 is (AIP) * (1/40) * e^(-0/40) = AIP/40 * 1
> quarter 1 is (AIP) * (1/40) * e^(-1/40) = AIP/40 * 0.97
> quarter 2 is (AIP) * (1/40) * e^(-2/40) = AIP/40 * 0.95
> ....   
>
> When space is returned, the AIP is calculated, the process starts over
> at year 0, quarter 0, etc.  I think ARIN staff will have a better view
> as to what timeframe to use conveniently.  If they can't decide on a
> timeframe, then additional guidance will be necessary.  Obviously, a
> very long timeframe is bad, for about the same reasons that extending
> the timeframe on delegation use is bad.  Too short a time frame would be 
> inconvenient, too.
>
>
>
>   
>>     * What mechanisms would be allowed to meet the needs of networks
>>       denied or delayed space under rationing?  Would a market be
>>       created/allowed such that networks that really need IP space right
>>       away can purchase it from other networks that can more easily free
>>       up addresses through improved efficiency?  Would networks needing
>>       space immediately be encouraged to get ("rent") PA space from a
>>       provider?
>>     
>
> I don't know about the whole market idea, yet.  A market allocates some
> resources quite effectively (e.g. Oil, Capital, Dry goods), and
> allocates some things quite badly (e.g. Healthcare, Law Enforcement,
> Fire Protection, National Security). I haven't quite decided for myself
> whether IP Addresses are like Oil or are like Healthcare.
>
> But I can't deny that a market of a sort exists now, and will probably
> exist whether or not you try to impose rules on prohibiting that.  
> Quite obviously, people with money who need IP Addresses can buy the
> ISPs that have already IP Addresses. Like the movie Wall Street, the
> raider can then turn the just-bought resources to their own more
> profitable purposes.  There is almost no way to stop that; because so
> far as ARIN is concerned, nothing has changed. The business pages just
> report ISP X bought ISP Y.  Allowing the sale of IP Address blocks on
> ebay would seem to make little difference.
>
> Recently, a person on a mailing list that I read, offered the use of his
> legacy /24 in exchange for hosting their server.  I can't see anything
> wrong with that.
>
> On the other hand, IP Address delegations are basically leases from the
> government, and the landlords (the government) can specify that the
> leases either are or aren't transferrable.  Usually, one wants some
> sensible cause for restrictions---e.g. you can't sell the nuclear
> weapons plant or whatever to just anyone---and so far in the "market"
> discussion, I haven't seen anyone really show the social harm in
> allowing blatant IP Address transferral for money.  I've only seen
> hypocritical discussion of what is fair and what limits should be
> imposed retroactively (and hypocritically) on others.  People speak of
> the probable USG interests; I think the USG will treat the issue of IP
> Addresses much like it treated frequency spectrum: Up for bid to the
> highest bidder, possibly subject to FCC regulation where necessary.
>
> But I do worry that a market for IP Addresses will eventually result in
> an Enron-like debacle with traders trying to cause power outages to
> generate higher prices and more profits.  Indeed, the Iraq war seems to
> have done somewhat the same thing in the Oil industry, but we don't have
> any tapes (yet) of Dick Cheney or anyone saying the equivalent of 'drop
> that plant off-line for maintenance during the heat wave' to spike
> prices like we do for Enron.  I've seen a lot (and have also
> occasionally been victim to)  operators and even senior people in the
> Internet whose morals are about the same or less than those of Enron
> traders.  But of course, the Enron, Worldcom, Adelphia, etc people were
> eventually found out and went to jail or were fired.
>
> If we create a market of IP Addresses, I think an Enron-like debacle is
> all but certain. So, the question for me is this: Is the benefit of a
> market worth the trouble?  Almost certainly there will be trouble.
> Almost certainly some people will go to jail on fraud, criminal
> conspiracy, etc like with Enron etc.  But outside the bad events, I
> think resources might be allocated quite well by a market.  But I don't
> know how damaging the bad events might be.  I'd guess they would be
> somehat similar to, but probably not as bad as, IP Address exhaustion.
>
>   
>>     * What do you mean by contested IP space?  Are you referring to
>>       pre-ARIN allocations and assignments, or something else?
>>     
>
> No. Just as I explained to Mr. Herrin, "contested" is a broad term to
> apply to ordinary disputes opened via the ticket system, and also to
> other disputes that may need to be negotiated or litigated.  This is
> motivated by the Kremen case.  It is so that ARIN staff can't ignore
> court orders, or otherwise pretend that they aren't subject to the
> jurisdiction of a court or otherwise ignore disputes.  I think this
> issue has to be written into policies in some way. The provision I wrote
> doesn't change anything else, but just prevents ARIN from considering
> contested space in its space available. That affects the hard limit
> function. The provision consequently ensures that should ARIN "lose" the
> ensuing negotiation or court case, there will be space available.
>
>   
>> I'm not sure if a rationing policy would be better than the Soft
>> Landing proposal, but IMO rationing is an idea worth fleshing out and
>> considering as an alternative.
>>     
>
> Thanks.
>
> 		--Dean
>
>