ietf
[Top] [All Lists]

Re: IPv6 address space shortages (was: Re: A simple question)

2003-04-29 14:28:26
Dean,

Since you have responded to a comment of mine, let me summarize my reasoning with regard to the "out of IPv4 address space" problem. Your assumptions, views, definitions, religion, etc., may differ.

--On Tuesday, 29 April, 2003 15:28 -0400 Dean Anderson <dean(_at_)av8(_dot_)com> wrote:

We are still not out of IPv4 address space.   I don't know the
projections for address space depletion,

One view of "address space depletion" is that projections are useless in an environment in which (i) the RIRs believe that their job is to prevent us from running out of space and (ii) they are good at their work. Those two statements imply that policies will adjust as they perceive we are getting closer to running out of space and, in turn, two seemingly-contradictory statements about the date on which we run out of IPv4 space:

        (1) We will _never_ run out of IPv4 space, but you
        really don't want to be the entity that asks for the
        next-to-last block (or any other block after your local
        RIR starts feeling significant end-game pressure).
        I.e., while we won't run out, getting space with get
        more and more expensive and/or aggravating, going up by
        step functions.
        
        (2) Given that model, it is more appropriate to look at
        the date at which it gets sufficiently difficult to get
        unique, globally-routable space (note that I did not say
        "PI") that people with a legitimate requirement for it
        perceive themselves as being unable to get it, or enough
        of it to meet their requirements.  Even 1918 recognizes
        that there are hosts that should have public addresses.
        To the extent that policies about IPv4 allocations are
        forcing such hosts behind NATs --or substitutes the
        judgment of RIRs or ISPs about what should, and should
        not, be public for that of local network
        managers/designers, or make it economically implausible
        to have public hosts-- then we have already passed that
        date.  And, to oversimplify things somewhat, it is
        accurate to say that we have already run out of space,
        because a rational criterion for "run out of space" is
        that it forces us into architecturally pathological
        behavior.


but I think the IPv6
problem won't be so much address space depletion as route
space depletion. I think it will be unwieldy to have a million
or 2 million or 10 million distinct routes.

Many of us agree with you, at least within the constraints of the current routing regime and models. But the question, on one side, is how to organize IPv6 allocations so that we don't have that problem. One theory is "almost no PI space except in very large aggregates, multiple (provider-dependent) addresses per host. As far as I can tell, that is the path we are going down. But it involves a number of tradeoffs which have, perhaps, not been explored and articulated as well as is desirable. One of them is that the strategy uses up addresses faster than the network grows (in host counts), even while it may conserve routing space. If you can anticipate the possibility of an RIR eventually telling an ISP "you can't count networks in your request for a larger allocation if they already have addresses from another provider", then we are in danger of being "out of" IPv6 address space coming out of the gate -- and that is one example of the motivation behind my challenge to Bill... and perhaps even of his "yet" comment which led to my challenge.

Aggregation won't
happen the way things are currently organized.

Maybe.  See above.

<soapbox> I think the RIRs (at least for IPv6) need to be
organized differently. In order for address space to be
hierachical, for efficient aggregation and routing, addresses
have to be given out locally, like phone numbers.
Unfortunately, that isn't how ISP's or the address registries
are organized. Fees are strongly biased towards big companies,
yet the small and medium sized companies generate most of the
revenue. It costs way too much money to get IPv6 space for
every town to have their own address space, so that even barn,
cow, and coke machine can be connected to the internet.
Something needs to give.  </soapbox>

Your conclusion may be right, but I don't think the reasoning in the paragraph above works. Let me restate it to see if I can explain why.

The only way we know for address space to be efficiently aggregated and routed is to make it hierarchical, and for that hierarchy to reflect routing realities. Unless we either change the business models of the industry, or create the type of regulatory regime that characterized the PSTN a few decades ago (which may be just one form of changing the business models), the hierarchy has to follow provider topologies and interconnection structures.

It makes sense for every town to have its own address space only if every town has a sufficient number of exchange/ interconnection points with other towns, or backbone providers of one sort or another to give the town global connectivity. But, in the general case, that is going to require a regulatory regime -- it isn't obviously in the interest of the ISPs, either technologically or economically, to make things work that way.

Of course, "address space per country" and, typically, address space per entity within country, is exactly the way address allocations work in the PSTN under the E.164 Recommendation/ System. But, if that type of address structure is to be used for routing, it depends on a collection of bilateral agreements (and exchange points) between countries, plus local and national routing regimes that are traditionally highly regulated (today's "deregulated" environment depends on the previous regulatory structure in a number of important ways), involve specialized and sometimes subsidized exchange points, small numbers of interexchange carriers relative to the number of exchanges (or event "towns"), etc.

I don't think we want to go there, but your opinions may, again, differ.

    john




<Prev in Thread] Current Thread [Next in Thread>