(apologies in advance for extending this thread here rather than on ppml -- will gladly take responses off-list, or move it over if responders would prefer to continue the discussion there) On Feb 22, 2008, at 6:22 AM, Stephen Sprunk wrote:

Thus spake "Adrian Chadd" <adrian@creative.net.au>

...

As I ranted on #nanog last night; the v6 transition will happen when it costs more to buy / maintain a v4 infrastructure (IP trading, quadruple NAT, support overheads, v6 tunnel brokers, etc) then it is to migrate infrastructure to v6.

If people were sane (!), they'd have a method right now for an enterprise to migrate 100% native IPv6 and interconnect to the v4 network via translation devices. None of this dual stack crap. It makes the heads of IT security and technical managers spin.

I agree, to a point. My prediction is that when the handful of mega-ISPs are unable to get the massive quantities of IPv4 addresses they need (a few dozen account for 90% of all consumption in the ARIN region)...

I keep reading assertions like this. Is there any public, authoritative evidence to support this claim? If there is, is this 90% figure a new development, or rather the product of changes in ownership (e.g., MCI-VZ-UU, SBC-ATT, etc.), changes in behavior (a run on the bank), some combination of the two, or something else altogether? Thanks, TV

dear arin hostfolk. could we please have the histogram for the last few years where the Y axis is the amount of allocation and the X axis is the number of organizations with that total size of new allocations during the period? you'll have to bucket alloc size in some useful way, probably a /16 or shorter or something. thanks. randy

ARIN has produced the histogram as requested and posted it to our website. It can be found at http://www.arin.net/statistics/index.html#ipv4org

leslie, thank you ever so much. but the way it depects the date kinda obscures my point. my apologies for being a pita, but could the y axis please be normalized to /24 or /32 equivalents, i.e. the amount of address space? thank you! randy

Operational comment: Look on the bright side, they may follow Comcast's example and deploy ipv6 instead!

Or they may not, and their share price will suffer as a result. People making the technical decision to stick with IPv4 for their large network are also making a decision to limit the growth of the network and to limit the growth of the business. As the IPv4 exhaustion issue becomes more widely understood, companies who have not prepared themselves to deploy IPv6 will find themselves under increasing scrutiny by shareholders. Comcast moved to IPv6 because their network was running out of RFC 1918 space. Since DOCSIS 3 includes IPv6 support, they made the decision to go to IPv6 rather than continue to spend money on shoehorning themselves into the limited IPv4 address space. Many people have not yet come to terms with how big the IPv6 space is, even the /32 that an ISP gets or the /48 that a site gets. We probably need to start talking about the number of subnetting bits available. For instance, an IPv4 ISP who assigns a /24 to subnets within their architecture and who has a /16 allocated for their architecture, has 8 bits available to subnet with. If an IPv6 ISP decides to assign a /64 to subnets and allocate a single /48 then they will also have 8 subnet bits available. So you could consider a /48 to be roughly equivalent to an IPv4 /16. Now, if an IPv6 ISP decides to strictly follow the rule of assigning a /48 per site internally, then each PoP or data center will be allocated a /48 meaning that each PoP or data center now has 8 subnet bits available. This amount of legroom allows you to do things like standardize subnet layouts for all sites, regardless of size, including the actual bits used from the 8 subnet bits. For instance, you can predict that if a PoP has 2001:1918:123/48 you know that if there is a switch connecting to a data center at that site, it will have the IPv6 address 2001:1918:123:d033::1 because your standard design has ::d033/64 assigned to the switch filling that role and Interface ID 1 assigned to its management interface. This kind of standardization makes it much easier to deploy PoPs regardless of whether it is in Dubai, where the data center is a half rack of webservers, or The Dalles where it is a 40,000 square foot warehouse. It also simplifies management and troubleshooting of the network. --Michael Dillon

In article <alpine.LSU.1.00.0802221545440.14814@hermes-1.csi.cam.ac.uk>, Tony Finch <dot@dotat.at> writes

...

I would not be surprised to learn that "consumption in the ARIN region" includes all the legacy assignments.

Many legacy assignments are now administered by the other RIRs http://www.iana.org/assignments/ipv4-address-space

I should have said: "...includes all the legacy assignments in the ARIN region". -- Roland Perry

...

So the quoted metric may well be true, but as unhelpful as claiming that "MIT has more address space than the whole of China" (as some people do from time to time).

Which is complete nonsense. MIT has 18/8, which is a little under 17 million addresses. I'm assuming that whatever else on top of that they have doesn't amount to a significant number. China is eating up IPv4 address space like it's going out of style (hm...) and they're now the third largest holder with 140 million IPv4 addresses, a hair shy of Japan's 142 million and 1/10th of the US's 1411 million.

Total delegations: 1624, millions of addresses: 48.55.

If one were to sum this up briefly, would it be correct to answer the MIT myth by saying: MIT has only 17 million addresses but China has 140 million. Along with Japan at 142 million, these are the top two holders of IP addresses with the USA trailing at 48.5 million. Due to legacy allocations which are often used wastefully due to legacy technology, the USA is often quoted as having 1,411 million IP addresses but this does not reflect the current rules under which IP address registries operate. In addition, since we are likely to use up all possible IPv4 addresses by 2011, smart organizations are moving to IPv6 where there is no shortage forecast for 100 years or more. Personally, I would like to see the NRO take a crack at issuing some kind of statement like this, to make it clear where the IP addresses are used, why organizations like MIT are not villains, and why the only way out of the steadily tightening straitjacket is to shift new network growth onto IPv6 and get to work on sorting out all the minor technical issues that will only get sorted out by actually pushing ahead with deployment, and use of IPv6. Back in the early days of the Internet, it was easier because there was a smaller community of vendors, network operators and protocol developers. Also, people didn't fully understand the implications of deploying the Internet as a replacement for all other networks in existence, therefore they forged ahead blissfully unaware that they were about to stumble head on into a technical problem. The result, is that there was constant movement, constant bug fixing, and all the minor technical issues faded into distant memories. All we have to do to make IPv6 ready for primetime is to deploy it for real, find the issues, fix the issues and move on. There is a real opportunity here for smaller companies who know how to run a lean mean operation, to deploy IPv6 Internet services at half the price of the large companies, and come out on top in three years or so when the large companies buy them out for big sums of money. The price incentive will ensure a steady stream of customers who are willing to take the chance with a less-than-perfect best-effort service. --Michael Dillon --Michael Dillon

...

If one were to sum this up briefly, would it be correct to answer the MIT myth by saying:

...

MIT has only 17 million addresses but China has 140 million. Along with Japan at 142 million, these are the top two holders of IP addresses with the USA trailing at 48.5 million.

Huh? Where do you get 48.5 million?

Well, your last posting was long and complicated and filled with figures but the second line from the bottom seemed to be saying that.

So the US has AT LEAST 278 million non-legacy addresses allocated/ assigned.

Fine. The point is for someone, (maybe you?) to do the research, sort out these figures, and provide us with a paragraph or so that amounts to a nice comprehensive sound-bite of the current state of affairs.

Also, some of the legacy blocks, such as 4/8 and 12/8 are de facto used by ISPs to address customers.

They are often used in a wasteful fashion because when the network architecture and subnetting was designed, there was no need to be frugal. Therefore, the big numbers are misleading. I was hoping to see something that is closer to what actual usage is.

Whichever way you slice it, the US is the largest holder of address space by a factor of more than 2,

No argument about that. But lots of it is inefficiently used with no way to recover the waste before we run out of IPv4. Water under the bridge.

and, until 2007, the largest user of new address space.

I wonder if that includes all the addresses that we registered with ARIN but use in Asia and Europe. This probably needs some serious research effort to bottom out. Not to mention all those companies, including two that I have worked for, who provided Internet access to European employees over access links in the USA. I think this is still rather common.

...

In addition, since we are likely to use up all possible IPv4 addresses by 2011,

No, that's when the depletion of the IANA pool is predicted. The RIRs also hold 400 million addresses for their day-to-day operations, which will take at least another year, maybe two, to deplete.

Assuming that there isn't a run on the bank. I expect that it will all disappear much quicker than we think. You probably noted that the UK had a real run on a real bank, last autumn. This is something that hadn't happened since the 19th century and I believe that in most of the USA, runs on the bank haven't been seen since the 1930's. This is primarily because people, through the last half of the 20th century, have TRUSTED their banks. But we all know that people, in general, have been growing less trustfull of governments, institutions, companies, etc. The decision makers at telecom companies are not likely to trust the RIRs or anyone else, and once the news of the IPv4 shortage filters into boardrooms, the rate of allocation will ramp up. Linear projections go out the window. Any kind of curve-fitting based on historical data goes out the window. The only way to truly forecast the demand for IPv4 addresses is to look at how vulnerable telecoms companies are in the are of IPv6. How easily can they make money with IPv6. How soon can IPv6 services become their core business. I'm afraid that everything we see shows that few telecoms companies are in a position to make IPv6 services their cash cow in 2011-2012. This means that IPv4 demand is going to go up, way up, as everybody scrambles to make sure that they have enough to carry them through the next two-three years. --Michael Dillon

On Feb 22, 2008, at 7:54 PM, Iljitsch van Beijnum wrote:

On 22 feb 2008, at 0:55, Tom Vest wrote:

...

...

I agree, to a point. My prediction is that when the handful of mega-ISPs are unable to get the massive quantities of IPv4 addresses they need (a few dozen account for 90% of all consumption in the ARIN region)...

...

I keep reading assertions like this. Is there any public, authoritative evidence to support this claim?

You can download files with all the delegation info from ftp.arin.net.

You mean the stats files, which provide delegation date, type, starting number, length, etc.? Which one of the published fields is the key field that enables you to identify the common recipient(s) of successive delegations over time? I'm assuming that the quoted 90% figure is some kind of aggregate (anything else would be pretty arbitrary), but I don't see anything in the public record that suggests how that aggregate might be produced...?

...

If there is, is this 90% figure a new development, or rather the product of changes in ownership (e.g., MCI-VZ-UU, SBC-ATT, etc.), changes in behavior (a run on the bank), some combination of the two, or something else altogether?

No, simply because large ISPs need lots of addresses, everyone else can make do with just a few.

But in the absence of some other metric for largeness, that sounds like a tautology. Large ISPs are the ones that demand lots of addresses... ergo to demand a lot of addresses is to be large... My question is not an entirely uninformed one. I'm quite familiar with the public stats. I just don't see how they transparently support this claim. Clarification would be greatly appreciated, TV

On 22 feb 2008, at 10:24, Roland Perry wrote:

...

I would not be surprised to learn that "consumption in the ARIN region" includes all the legacy assignments.

By definition, no new legacy assignments are given out. :-)

So simply looking at recent data will correct for this.

...

So the quoted metric may well be true, but as unhelpful as claiming that "MIT has more address space than the whole of China" (as some people do from time to time).

Which is complete nonsense. MIT has 18/8, which is a little under 17 million addresses. I'm assuming that whatever else on top of that they have doesn't amount to a significant number. China is eating up IPv4 address space like it's going out of style (hm...) and they're now the third largest holder with 140 million IPv4 addresses, a hair shy of Japan's 142 million and 1/10th of the US's 1411 million.

On 22 feb 2008, at 10:31, Randy Bush wrote:

...

dear arin hostfolk. could we please have the histogram for the last few years where the Y axis is the amount of allocation and the X axis is the number of organizations with that total size of new allocations during the period? you'll have to bucket alloc size in some useful way, probably a /16 or shorter or something.

I can't see organizations in ARIN's delegation records, but simply counting delegations and rounding sizes to the closest power of 2 results this for 20070101 - now:

+------+-------------+--------+ | size | delegations | Maddrs | +------+-------------+--------+ | 10 | 2 | 6.82 | | 11 | 5 | 11.27 | | 12 | 6 | 6.14 | | 13 | 6 | 2.96 | | 14 | 5 | 1.14 | | 15 | 12 | 1.58 | | 16 | 24 | 1.53 | | 17 | 27 | 0.87 | | 18 | 51 | 0.82 | | 19 | 110 | 0.90 | | 20 | 474 | 1.94 | | 21 | 227 | 0.46 | | 22 | 415 | 0.42 | | 23 | 1 | 0.00 | | 24 | 11 | 0.00 | +------+-------------+--------+

Totals:

+-------------+--------+ | delegations | Maddrs | +-------------+--------+ | 1376 | 36.86 | +-------------+--------+

I.e., /18 or shorter is 134 delegations (10%) and 33.08 million addresses (90%).

However, ARIN has the unfortunate practice of backdating delegations when people come back for more address space and the new and old blocks can form a bigger block. Below the same numbers but with logic that tries to correct for this, which makes it impossible to easily show the correct numbers of delegations and addresses in one table:

+------+-------------+ | size | delegations | +------+-------------+ | 8 | 1 | | 10 | 4 | | 11 | 13 | | 12 | 12 | | 13 | 12 | | 14 | 17 | | 15 | 35 | | 16 | 38 | | 17 | 61 | | 18 | 95 | | 19 | 222 | | 20 | 440 | | 21 | 231 | | 22 | 425 | | 23 | 5 | | 24 | 13 | +------+-------------+

+------+--------+ | size | Maddrs | +------+--------+ | 8 | 3.15 | | 10 | 7.34 | | 11 | 16.58 | | 12 | 8.37 | | 13 | 2.74 | | 14 | 1.39 | | 15 | 3.31 | | 16 | 0.14 | | 17 | 1.12 | | 18 | 0.84 | | 19 | 1.39 | | 20 | 1.27 | | 21 | 0.47 | | 22 | 0.43 | | 23 | 0.00 | | 24 | 0.00 | +------+--------+

Total delegations: 1624, millions of addresses: 48.55.

/18 or more: 195 (12%), 44.16 (91%).

Hi Iljitsch, Thanks for your response. On Feb 23, 2008, at 1:38 AM, Iljitsch van Beijnum wrote:

On 22 feb 2008, at 16:41, Tom Vest wrote:

...

...

You can download files with all the delegation info from ftp.arin.net.

...

You mean the stats files, which provide delegation date, type, starting number, length, etc.?

Yes.

...

Which one of the published fields is the key field that enables you to identify the common recipient(s) of successive delegations over time?

There is no such field.

I didn't think so. So there is no accurate way to get anything like a sum of IP address per LIR at any point in time, now or in the the past, at least not using publicly available data. Given that impossibility, I still don't see how anyone can make the (increasingly oft repeated) claim that 90% (or any specific share) of address space is now going to some subset of the LIRs... no?

...

...

No, simply because large ISPs need lots of addresses, everyone else can make do with just a few.

...

But in the absence of some other metric for largeness, that sounds like a tautology. Large ISPs are the ones that demand lots of addresses... ergo to demand a lot of addresses is to be large...

You've got a point there. However, I think many of us will be able to judge ISP size from other factors and observe that the correlation by the such determined ISP size and address use is quite high.

I agree that many of us can estimate ISP size even more accurately, by looking at the sum of address space originated by well-known ASes associated with those ISPs. I think many of us will recognize that there may be other, less well-known ASes associated with some of these, and so an accounting of the well-known ones is incomplete, perhaps a lower bound. I agree that some of us can correlate the contents of the routing table over time with the entries in the delegated files, to get very loose inter-temporal (delegation- origination) associations, which have been shaped over time in opaque ways by M&A, multihoming, customer management and traffic engineering engineering practices, etc. However, I still haven't seen anything that enables one to penetrate this fog of largely unknowable commercial and operational details sufficiently to justify the 90% claim -- or any other claim. If there is some known method for doing this, and hence some defensible way to derive the actual (maybe 90%?) ratio, then I'd still be very interested to hear about it! I think all of the academics who spent several years trying (with mixed results) to come up with algorithms for inferring inter-AS relationships, etc. would be very interested too! Thanks, TV

In article <A088046F-DA64-4676-9C7B-5313F9145A3C@muada.com>, Iljitsch van Beijnum <iljitsch@muada.com> writes

I'm not sure why exactly you want to know how much space goes to how many organizations

Several days ago, it seemed to me that Stephen Sprunk suggested that it would only take a change of policy of a handful of large ISPs (I'm carefully using new words here), to think "party's over" and start converting their users to 10/8 addresses, and therefore 90% of the demand for new allocations dries up. On the other hand, if the 90% of allocations are going to (large) new entrants, and others with a less homogenous or convertible user base, the demand might not dry up so suddenly.

We know that pretty much 10% of the requests is responsible for 90% of the address space. So apparently 90% of the address space is going to at most 10% of the LIRs.

What we haven't established yet is whether this is the same 10% that already had 90% of the allocations (from last century), growing their empire, or new kids on the block. -- Roland Perry

On Feb 23, 2008, at 1:54 PM, Stephen Sprunk wrote:

Thus spake "Tom Vest" <tvest@eyeconomics.com>

...

...

I agree, to a point. My prediction is that when the handful of mega-ISPs are unable to get the massive quantities of IPv4 addresses they need (a few dozen account for 90% of all consumption in the ARIN region)...

I keep reading assertions like this. Is there any public, authoritative evidence to support this claim?

Rechecking my own post to PPML, 73 Xtra Large orgs held 79.28% of ARIN's address space as of May 07; my apology for a faulty memory, but it's not off by enough to invalidate the point.

The statistics came from ARIN Member Services in response to an email inquiry. I don't believe they publish such things anywhere (other than what's in WHOIS), but you can verify yourself if you wish; they were quite willing to give me any stats I asked for if they had the necessary data available.

Thanks for the information Stephen. In order to be perfectly clear on how to interpret this, it would be good to know whether this sum includes the pre-ARIN delegations, or just reflects what has happened since ARIN was established. Then, if the distribution is deemed to be of real significance, e.g., to understand the past, or better understand what to do next, the best/simplest metric for such evaluations is probably the Herfindahl- Hirschman Index (HHI), which attempts to capture both dimensions of concentration/diffusion in a single unidimensional scale. This is probably the most common tool that economists and policymakers use to evaluate such things when necessary (e.g., before approving very large mergers) in other sectors. The standard interpretations for HHI results may not be ideal for this sector, but if one has access to time series data, it does provide a really nice way of evaluating concentration/diffusion trends over time. To my mind, concerns about the distribution of PA are pretty analogous to concerns about the distribution of access to goods in general, while PA+PI together could be interpreted as the distribution of final goods themselves (e.g., think airports or airlines vs. individual passenger-flights). Since we're not talking about goods that are strongly bound to geography, the fact that ARIN members don't all "compete" for the same kinds of customers or over the same market territories doesn't really matter, as it would in more conventional contexts. More on HHI here: http://en.wikipedia.org/wiki/Herfindahl_index Bottom line: if people believe that the kind of de facto concentration/diffusion claims that have been circulating are important, a bit more effort should probably be devoted to understanding how to sensibly interpret such things. TV

...

If there is, is this 90% figure a new development, or rather the product of changes in ownership (e.g., MCI-VZ-UU, SBC-ATT, etc.), changes in behavior (a run on the bank), some combination of the two, or something else altogether?

Most of the orgs in the Xtra Large class were already there before the mega-mergers started; after all, you only need >/14 to be Xtra Large. Given how most tend to operate in silos, they might still be separate orgs as far as ARIN is concerned...

S

Stephen Sprunk "God does not play dice." --Albert Einstein CCIE #3723 "God is an inveterate gambler, and He throws the K5SSS dice at every possible opportunity." --Stephen Hawking

On Feb 24, 2008, at 12:45 PM, Stephen Sprunk wrote:

Thus spake "Tom Vest" <tvest@eyeconomics.com>

...

On Feb 23, 2008, at 1:54 PM, Stephen Sprunk wrote:

...

Rechecking my own post to PPML, 73 Xtra Large orgs held 79.28% of ARIN's address space as of May 07; my apology for a faulty memory, but it's not off by enough to invalidate the point.

The statistics came from ARIN Member Services in response to an email inquiry. I don't believe they publish such things anywhere (other than what's in WHOIS), but you can verify yourself if you wish; they were quite willing to give me any stats I asked for if they had the necessary data available.

Thanks for the information Stephen. In order to be perfectly clear on how to interpret this, it would be good to know whether this sum includes the pre-ARIN delegations, or just reflects what has happened since ARIN was established.

The wording of the question and response referred only to "ARIN members". That does not include most orgs with _only_ legacy allocations, but it would include orgs with both legacy and non- legacy allocations. Presumably, if an org had both types, both would have been included, but that wasn't explicitly stated since it wasn't relevant to the questions I was asking at the time.

Not necessarily. Orgs which are end-users and not LIR/ISP subscriber members may have resources from ARIN without being members. Owen

Stephen Sprunk wrote:

Thus spake "Tom Vest" <tvest@eyeconomics.com>

...

...

I agree, to a point. My prediction is that when the handful of mega-ISPs are unable to get the massive quantities of IPv4 addresses they need (a few dozen account for 90% of all consumption in the ARIN region)...

I keep reading assertions like this. Is there any public, authoritative evidence to support this claim?

Rechecking my own post to PPML, 73 Xtra Large orgs held 79.28% of ARIN's address space as of May 07; my apology for a faulty memory, but it's not off by enough to invalidate the point.

I can phrase it another way, working solely off the published data from the 5 RIRs (daily stats files) Of the 13,521 IPv4 allocation transactions since 1 January 2006, some 103 transactions accounted for 42% of the total volume of IPv4 address space allocated in this period. (i.e. an allocation event of a /12 or greater) 24 were recorded in the APNIC report, 37 in the ARIN report(*) and 31 in the RIPE report

The statistics came from ARIN Member Services in response to an email inquiry. I don't believe they publish such things anywhere (other than what's in WHOIS), but you can verify yourself if you wish; they were quite willing to give me any stats I asked for if they had the necessary data available.

...

If there is, is this 90% figure a new development, or rather the product of changes in ownership (e.g., MCI-VZ-UU, SBC-ATT, etc.), changes in behavior (a run on the bank), some combination of the two, or something else altogether?

Most of the orgs in the Xtra Large class were already there before the mega-mergers started; after all, you only need >/14 to be Xtra Large. Given how most tend to operate in silos, they might still be separate orgs as far as ARIN is concerned...

This data regarding allocations does not reflect after-the-event mergers. It simply looks at the size distribution in the daily stats files as reported by the RIRs. In ARIN's case in particular 57% of all IPv4 addresses allocated since 1 January 2006 were allocated as part of a /11 or larger, and 88% were part of a /16 or greater. This equates to 6 transactions of a /11 (out of 3,546 individual transactions for the same period, or fractions of a percent. For /16 or larger there were 306 transactions out of 3,546, or 8.6%. Thats more than "a few dozen", but it does not also reflect mergers and aquisitions post allocation. (*The ARIN report format had to be re-processed becuase of the differing procedure ARIN uses to update this report each day) Geoff