[arin-ppml] DRAFT POLICY 2012-3: ASN TRANSFERS

[Tom Vest]

On Mar 30, 2012, at 5:39 PM, David Farmer wrote:

> 
> 
> On 3/30/12 11:54 CDT, Scott Leibrand wrote:
>> On Mar 30, 2012, at 9:28 AM, Michael Sinatra<michael+ppml at burnttofu.net>  wrote:
>>> You seem to be saying that it's not actual reputation that is being
>>> traded, but some sort of misperception that ASNs in a certain number
>>> range bring credibility.  So rather than trade on actual reputation
>>> (which I would question in itself), you are advocating creating a market
>>> where a fake perception of reputation is what's traded.  That doesn't
>>> sound to me like a market that will efficiently allocate resources,
>>> although it may efficiently allocate misperception.
>>> 
>>> IPv4 number transfers make sense.  IPv6 number transfers do not.  I am
>>> on the fence about ASN transfers, but it's arguments like these in favor
>>> that are making me increasingly wary.
>> 
>> IMO it's not about reputation, it's about ease of use. A shorter ASN is easier to remember, say, and recognize. When I was setting up peering at a former job, we needed a new ASN for the peering network, and had a few unused ASNs to choose from. We chose the one that was easiest for humans (22212). If we could've easily acquired a 3-digit ASN instead, that would've been even better.
>> 
>> There are fewer than 9999 companies doing peering at IXs, and likely fewer than 999 that have more than a few dozen peers. IMO there's no good reason any of them should have to use a hard-to-remember random 5-digit ASN for peering if they don't want to.
>> 
>> -Scott
> 
> This Human Factors based argument makes sense to me.  It is on par with making IPv6 allocations on nibble boundaries and the fact that IPv6 has zero suppression, because they also makes things easier for Humans. Until the Internet starts building itself, Human Factors are always going to be an issue to some degree or another.
> 
> Number are just numbers to computers.  However, when Humans interact with the numbers as part of the system, smaller or easy to remember longer sequences will be better and less likely to cause transcription or other Human based errors.

Hi David, 

Please consider the following two counterarguments.

First, while the "human factors" interest does indeed merit consideration, the relative ease of remembering a five-digit vs. ten-digit number is hardly the only or even the most important interest at stake in how ASes are distributed. ISTM that a routed prefix' is ultimately operationally "authenticated" by means of/reference to its origin-AS (though this may have been easy for many to forget during the last decade of non-transferable IPv4 and rapidly proliferating single prefix-originating singlehomed ASes) -- and the ASes themselves are currently "authenticated" through the RIR needs-based allocation procedure. Sure, today some ASNs are short and easier both to remember and configure, while others are long and give some people discomfort -- but changing the mechanisms through which they're distributed would affect much more than just the quantity of "easy numbers" available to the AS-seeking population.

As an analogy, one might consider the relationship between automobiles and license plates in the United States. The contents of the "authentication" embodied in different license tokens (plates, stickers, etc.) may vary from state to state, but they all play some kind of role in attesting not just to the "uniqueness" of a car, but also to the unique relationship between each vehicle and the party that bears ultimate responsibility for how it is used. Basically that's what "authentication" means. A trip down any US street clearly reveals both that "human factors" influence people's tastes in this numbering domain as well, and also that the some kind of mechanism exists to accommodate those differences in tastes. Everywhere, demand seems to be especially high for certain kinds of license strings (esp. anything related to the home-team), and no doubt there are many people who might be willing to pay a large premium to obtain an especially pithy license string from whomever currently possesses it. But the question is: even assuming that you are one of those persons who would be willing to pay a lot to acquire someone else's vanity license plate without actually acquiring the attached car, would you be better off *as a driver* if the law allowed you to buy and sell license plates at will with anyone at any time for any reason, so long as the other party is willing to strike a deal? Apart from simplifying your quest for a license string that that uniquely satisfies your own personal preferences, can you imagine how else might that change affect your life as a driver -- e.g., incidence of auto-related theft and other crimes, cost and availability of insurance, etc. etc.? (Note: bonus points if you can further imagine how this situation might differ if the cars in question were just as capable of causing damage as conventional autos are today, but in additional were invisible, ephemeral, and could appear and/or disappear at will anywhere in the world, like ASNs...) 

Second, while again the "human factors" interest does indeed deserve serious consideration -- esp. during a technology's design phase -- when the context is market rather than technology design, it's important to distinguish between factors that might make certain market alternative(s) more or less useful/valuable to some market participants but not others -- i.e., about which opinions can vary both across individuals and even within individual market participants over time (as people change their minds) -- vs. factors that would have the same kind of effect but universally, i.e., which would cause *all* market participants to agree that one kind of factor is absolutely more valuable/useful or "superior" to any alternative. I'd argue that this distinction is especially important to recognize and understand when considering network-related markets, and double-plus important to take into account during times of "market transition" (such as the one we'll undergo as the ASN16 pool is exhausted), because at such moments a flawed market might cause the former kind of variable/transient preferences (i.e., "personal tastes") to harden into the latter sort of uniform judgment, causing the absolute rejection/market failure/technical non-viability of the less favored alternative. 

If that seems too abstract, try out this example. How do you feel about CIDR today? As a veteran of the classful networking days, do you feel that it took more effort to learn CIDR and more time/effort/$$$ to actually transition to CIDR (and IDR in general) than it would have to just continue on as before with classful networking forever -- *if* that option had been available to you? Do you think that, instead of migrating directly to CIDR, things would have turned out better if the Class A and B holders of the early 1990s had held off on CIDR adoption for a while in order to try out IPv4 privatization and see how high aspiring customers would bid up the price of ever-scarcer classful prefixes? Since the value of anyone else adopting CIDR *before* the largest providers of the time had signaled their absolute commitment to go that way would have been marginal if not negative, how do you think that minor accommodation of "human factors" would have affected the timing of CIDR adoption? Would the net effect of that minor change have been positive or negative, and for who?

In the end you may still feel that the benefits of embracing ASN transfers outweigh the dangers; I just ask that you recognize and consider both.

Thanks, 

TV