Alternative Re: ipv4/25s and above Re: 202211232221.AYC

Vasilenko Eduard vasilenko.eduard at huawei.com
Mon Nov 28 08:00:26 UTC 2022


Big OTTs installed caches all over the world.
Big OTTs support IPv6.
Hosts prefer IPv6.
Hence, traffic becomes IPv6 to big OTTs.
It is not visible for IXes. IXes statistics on IPv6 are not representative.
Ed/
-----Original Message-----
From: Abraham Y. Chen [mailto:aychen at avinta.com] 
Sent: Sunday, November 27, 2022 12:35 AM
To: Chris Welti <chris.welti at switch.ch>
Cc: NANOG <nanog at nanog.org>; bzs at theworld.com; Vasilenko Eduard <vasilenko.eduard at huawei.com>
Subject: Re: Alternative Re: ipv4/25s and above Re: 202211232221.AYC

Hi, Chris:

1) "... public fabric ... private dedicated circuits ... heavily biased
...":   You brought up an aspect that I have no knowledge about. 
However, you did not clarify how IPv6 and IPv4 are treated differently by these considerations which was the key parameter that we are trying to sort out. Thanks.

Regards,

Abe (2022-11-24 15:40)


On 2022-11-24 12:23, Chris Welti wrote:
> Hi Abe,
>
> the problem is that the AMS-IX data only covers the public fabric, but 
> the peering connections between the big CDNs/clouds and the large ISPs 
> all happen on private dedicated circuits as it is so much traffic that 
> it does not make sense to run it over a public IX fabric (in addition 
> to local caches which dillute the stats even more). Thus that data you 
> are referring to is heavily biased and should not be used for this 
> generalized purpose.
>
> Regards,
> Chris
>
> On 24.11.22 18:01, Abraham Y. Chen wrote:
>> Hi, Eduard:
>>
>> 0) Thanks for sharing your research efforts.
>>
>> 1) Similar as your own experience, we also recognized the granularity 
>> issue of the data in this particular type of statistics. Any data 
>> that is based on a limited number of countries, regions, businesses, 
>> industry segments, etc. will always be rebutted with a counter 
>> example of some sort. So, we put more trust into those general 
>> service cases with continuous reports for consistency, such as 
>> AMS-IX. If you know any better sources, I would like to look into them.
>>
>> Regards,
>>
>>
>> Abe (2022-11-24 11:59 EST)
>>
>>
>> On 2022-11-24 04:43, Vasilenko Eduard wrote:
>>> Hi Abraham,
>>> Let me clarify a little bit on statistics - I did an investigation 
>>> last year.
>>>
>>> Google and APNIC report very similar numbers. APNIC permits drilling 
>>> down deep details. Then it is possible to understand that they see 
>>> only 100M Chinese. China itself reports 0.5B IPv6 users. APNIC gives 
>>> Internet population by country - it permits to construct proportion.
>>> Hence, it is possible to conclude that we need to add 8% to Google 
>>> (or APNIC) to get 48% of IPv6 preferred users worldwide. We would 
>>> likely cross 50% this year.
>>>
>>> I spent a decent time finding traffic statics. I have found one DPI 
>>> vendor who has it. Unfortunately, they sell it for money.
>>> ARCEP has got it for France and published it in their "Barometer". 
>>> Almost 70% of application requests are possible to serve from IPv6.
>>> Hence, 70%*48%=33.6%. We could claim that 1/3 of the traffic is IPv6 
>>> worldwide because France is typical.
>>> My boss told me "No-No" for this logic. His example is China where 
>>> we had reliable data for only 20% of application requests served on
>>> IPv6 (China has a very low IPv6 adoption by OTTs).
>>> My response was: But India has a much better IPv6 adoption on the 
>>> web server side. China and a few other countries are not 
>>> representative. The majority are like France.
>>> Unfortunately, we do not have per-country IPv6 adoption on the web 
>>> server side.
>>> OK. We could estimate 60% of the application readiness as a minimum. 
>>> Then 60%*48%=28.8%.
>>> Hence, we could claim that at least 1/4 of the worldwide traffic is 
>>> IPv6.
>>>
>>> IX data shows much low IPv6 adoption because the biggest OTTs have 
>>> many caches installed directly on Carriers' sites.
>>>
>>> Sorry for not the exact science. But it is all that I have. It is 
>>> better than nothing.
>>>
>>> PS: 60% of requests served by web servers does not mean "60% of 
>>> servers". For servers themselves we have statistics - it is just 
>>> 20%+. But it is for the biggest web resources.
>>>
>>> Eduard
>>> -----Original Message-----
>>> From: NANOG
>>> [mailto:nanog-bounces+vasilenko.eduard=huawei.com at nanog.org] On 
>>> Behalf Of Abraham Y. Chen
>>> Sent: Thursday, November 24, 2022 11:53 AM
>>> To: Joe Maimon<jmaimon at jmaimon.com>
>>> Cc: NANOG<nanog at nanog.org>;bzs at theworld.com
>>> Subject: Re: Alternative Re: ipv4/25s and above Re: 202211232221.AYC
>>>
>>> Dear Joe:
>>>
>>> 0) Allow me to share my understanding of the two topics that you 
>>> brought up.
>>>
>>> 1) "...https://www.google.com/intl/en/ipv6/statistics.html, it looks 
>>> like we’ve gone from ~0% to ~40% in 12 years.... ": Your numbers may 
>>> be deceiving.
>>>
>>>     A. The IPv6 was introduced in 1995-12, launched on 2012-06-06 
>>> and ratified on 2017-07-14. So, the IPv6 efforts have been quite a 
>>> few years more than your impression. That is, the IPv6 has been 
>>> around over quarter of a century.
>>>
>>>     B. If you read closely, the statement  "The graph shows the 
>>> percentage of users that access Google over IPv6." above the graph 
>>> actually means "equipment readiness". That is, how many Google users 
>>> have IPv6 capable devices. This is similar as the APNIC statistics 
>>> whose title makes this clearer. However, having the capability does 
>>> not mean the owners are actually using it. Also, this is not general 
>>> data, but within the Google environment. Since Google is one of the 
>>> stronger promoters of the IPv6, this graph would be at best the cap 
>>> of such data.
>>>
>>>     C. The more meaningful data would be the global IPv6 traffic 
>>> statistics. Interestingly, they do not exist upon our extensive search.
>>> (If you know of any, I would appreciate to receive a lead to such.) 
>>> The closest that we could find is % of IPv6 in AMS-IX traffic 
>>> statistics (see URL below). It is currently at about 5-6% and has 
>>> been tapering off to a growth of less than 0.1% per month recently, 
>>> after a ramp-up period in the past. (Similar saturation behavior can 
>>> also be found in the above Google graph.)
>>>
>>> https://stats.ams-ix.net/sflow/ether_type.html
>>>
>>>     D.  One interesting parameter behind the last one is that as an 
>>> Inter-eXchange operator, AMS-IX should see very similar percentage 
>>> traffic mix between IPv6 and IPv4. The low numbers from AMS-IX does 
>>> not support this viewpoint for matching with your observation. In 
>>> addition, traffic through IX is the overflow among backbone routers.
>>> A couple years ago, there was a report that peering arrangements 
>>> among backbone routers for IPv6 were much less matured then IPv4, 
>>> which meant that AMS-IX should be getting more IPv6 traffic than the 
>>> mix in the Internet core. Interpreted in reverse, % of IPv6 in 
>>> overall Internet traffic should be less than what AMS-IX handles.
>>>
>>>     E. This is a quite convoluted topic that we only scratched the 
>>> surface. They should not occupy the attention of colleagues on this 
>>> list. However, I am willing to provide more information to you 
>>> off-line, if you care for further discussion.
>>>
>>> 2) 
>>> "...https://lore.kernel.org/lkml/[email protected]/
>>> ...":  My basic training was in communication equipment hardware 
>>> design.
>>> I knew little about software beyond what I needed for my primary 
>>> assignment. Your example, however, reminds me of a programing course 
>>> that I took utilizing APL (A Programming Language) for circuit 
>>> analysis, optimization and synthesis. It was such a cryptic symbolic 
>>> language that classmates (mostly majored in EE hardware) were 
>>> murmuring to express their displeasure. One day we got a homework 
>>> assignment to do something relatively simple. Everyone struggled to 
>>> write the code to do the job.
>>> Although most of us did get working codes, they were pages long. The 
>>> shortest one was one full page. Upon reviewed all homework, the 
>>> professor smiled at us and told us to look for the solution section 
>>> at the end of the text book. It turned out to be the answer for a 
>>> problem in the next chapter to be covered. The code was only three 
>>> lines long!
>>> Although it did not have the codes for debugging purposes, it 
>>> covered all error messages expected. It was such a shocker that 
>>> everyone quieted down to focus on the subject for the rest of the 
>>> semester. During my first employment, we had the need to optimize 
>>> circuit designs. Since I was the only staff who knew about it, I 
>>> ended up being the coordinator between several hardware designers 
>>> and the supporting programmer. From that teaching, I am always 
>>> looking for the most concise solution to an issue, not being 
>>> distracted or discouraged by the manifestation on the surface.
>>>
>>> https://en.wikipedia.org/wiki/APL_(programming_language)
>>>
>>> 3) Fast forward half a century, I am hoping that my "one-line code"
>>> serves the purpose of "there exists" an example in proofing a 
>>> mathematical theorem for  inspiring software colleagues to review 
>>> the network codes in front of them for improvement, instead of 
>>> presenting such as a valid hurdle to progress.
>>>
>>>
>>> Regards,
>>>
>>>
>>> Abe (2022-11-24 03:53 EST)
>>>
>>>
>>>
>>>
>>>
>>> On 2022-11-21 19:30, Joe Maimon wrote:
>>>> David Conrad wrote:
>>>>> Barry,
>>>>>
>>>>> On Nov 21, 2022, at 3:01 PM,bzs at theworld.com wrote:
>>>>>> We've been trying to get people to adopt IPv6 widely for 30 years 
>>>>>> with very limited success
>>>>> According tohttps://www.google.com/intl/en/ipv6/statistics.html, 
>>>>> it looks like we’ve gone from ~0% to ~40% in 12 years.
>>>>> https://stats.labs.apnic.net/ipv6 has it around 30%. Given an 
>>>>> Internet population of about 5B, this can (simplistically and
>>>>> wrongly) argued to mean 1.5-2B people are using IPv6. For a 
>>>>> transition to a technology that the vast majority of people who 
>>>>> pay the bills will neither notice nor care about, and for which 
>>>>> the business case typically needs projection way past the normal 
>>>>> quarterly focus of shareholders, that seems pretty successful to me.
>>>>>
>>>>> But back to the latest proposal to rearrange deck chairs on the 
>>>>> IPv4 Titanic, the fundamental and obvious flaw is the assertion of 
>>>>> "commenting out one line code”. There isn’t “one line of code”. 
>>>>> There are literally _billions_ of instances of “one line of code”, 
>>>>> the vast majority of which need to be changed/deployed/tested with 
>>>>> absolutely no business case to do so that isn’t better met with 
>>>>> deploying
>>>>> IPv6+IPv4aaS. I believe this has been pointed out numerous times, 
>>>>> IPv6+but
>>>>> it falls on deaf ears, so the discussion gets a bit tedious.
>>>>>
>>>>> Regards,
>>>>> -drc
>>>>>
>>>> Had the titanic stayed afloat some hours more, many more would have 
>>>> survived and been rescued when assistance eventually arrived. So 
>>>> that makes this a debate over whether this is deck chair 
>>>> re-arrangement or something more meaningful.
>>>>
>>>> As I and others have pointed out, it depends on how it is used. And 
>>>> perhaps the attempt should be made regardless of knowing in advance 
>>>> which it will be.
>>>>
>>>> You assertion needs some back of the envelope numbers, which once 
>>>> provided, I suspect will render your estimate grossly incorrect.
>>>>
>>>> You can hardly attempt to convince anybody that 240/4 as unicast 
>>>> would not be the more trivial change made in any of these products 
>>>> natural life cycle points.
>>>>
>>>> Especially as we have examples of what that type of effort might 
>>>> look like. IGTFY and here
>>>>
>>>> https://lore.kernel.org/lkml/[email protected]/
>>>>
>>>> The burdensome position is ridiculous even more so when stated with 
>>>> a straight face.
>>>>
>>>> Joe
>>>>
>>>>
>>>>
>>> --
>>> This email has been checked for viruses by Avast antivirus software.
>>> www.avast.com
>>>
>>
>>



More information about the NANOG mailing list