IPv6 faster/better proof? was Re: Need /24 (arin) asap

Lee Howard lee.howard at retevia.net
Wed Jun 13 11:46:31 UTC 2018


On 06/11/2018 05:16 PM, Scott Weeks wrote:
>
> --- cb.list6 at gmail.com wrote:
> From: Ca By <cb.list6 at gmail.com>
>
>> Meanwhile, FB reports that 75% of mobiles in the USA
>> reach them via ipv6
>>
>> And Akaimai reports 80% of mobiles
> And they both report ipv6 is faster / better.
> ----------------------------------------
Let me grab a few more for you:

https://blogs.akamai.com/2016/06/preparing-for-ipv6-only-mobile-networks-why-and-how.html 


https://blogs.akamai.com/2016/10/ipv6-at-akamai-edge-2016.html

https://www.theregister.co.uk/2016/07/28/ipv6_now_faster_a_fifth_of_the_time 
which cites an academic paper 
http://dl.acm.org/citation.cfm?doid=2959424.2959429 by Vaibhav Bajpai 
and Jürgen Schönwälder

https://www.linkedin.com/pulse/ipv6-measurements-zaid-ali-kahn/

https://community.infoblox.com/t5/IPv6-CoE-Blog/Can-IPv6-Rally-Be-Faster-than-IPv4-Part-1/ba-p/6419 


https://www.nanog.org/meetings/abstract?id=2281

> I'd sure like to see how they came up with these
> numbers in a technically oriented paper.
Most of the above links explain how they got the numbers.
Facebook, in particular, did A/B testing using Mobile Proxygen, which is 
to say that they configured their mobile app to report performance over 
both IPv4 and IPv6 from the same handset at the same time.
Others, including APNIC's https://stats.labs.apnic.net/v6perf have a 
browser fetch two objects with unique URLs, one from an IPv4-only server 
and one from an IPv6-only server, and compare them.



>   There
> should be no difference, except for no CGN or Happy
> Eyeballs working better or something similar.  Am I
> missing something?  Same routers; same links; same
> RTTs; same interrupt times on servers; same etc, etc
> for both protocols.

 From time to time somebody says, "Okay, maybe it works in practice, but 
does it work in *theory*?"

Busy engineers hardly ever investigate things going inexplicably right.

My hypothesis is that the observed difference in performance relates to 
how mobile networks deploy their transition mechanisms. Those with a 
dual-stack APN take a native path for IPv6, while using a CGN path for 
IPv4, which, combined with the Happy Eyeballs head start, might add 
501microseconds, which is a ms, which is 15% of 7ms. Those with an 
IPv6-only APN use a native path for IPv6, while using either a NAT64 for 
IPv4 (identical performance to CGN) or 464xlat, which requires 
translation both in the handset and the NAT64; handsets may not be 
optimized for header translation.

However, I have a dozen other hypotheses, and the few experiments I've 
been able to run have not confirmed any hypothesis. For instance, when 
one protocol is faster than another on a landline network, hop count is 
not a correlation (therefore, shorter paths, traffic engineering, etc., 
are not involved).

Lee

>
> Hmm...  Faster and better?
>
> The links seem to be an IPv6 cheerleader write up.
> I looked at the URLs and the URLs one pointed to and
> pulled out everything related to IPv6 being
> faster/better.
>
>
> Akamai URL:
>
> "For dual-stacked hostnames we typically see higher
> average estimated throughput over IPv6 than over IPv4.
> Some of this may be due to IPv6-connected users being
> correlated with better connectivity, but over half of
> dual-stacked hostnames (weighted by daily bytes
> delivered) have IPv6 estimated throughput at least 50%
> faster than IPv4, and 90% of these hostnames have the
> IPv6 estimated throughput at least 10% faster than
> IPv4."
>
>
>
> FB URL:
>
> "People using Facebook services typically see better
> performance over IPv6..."
>
> and it points to
> https://code.facebook.com/posts/1192894270727351/ipv6-it-s-time-to-get-on-board
> which says:
>
> "We’ve long been anticipating the exhaustion of IPv
> in favor of the speed and performance benefits of
> IPv6."
>
> "We’ve observed that accessing Facebook can be 10-15
> percent faster over IPv6."
>
>
> I'd sure like to see how they came up with these
> numbers in a technically oriented paper.  There
> should be no difference, except for no CGN or Happy
> Eyeballs working better or something similar.  Am I
> missing something?  Same routers; same links; same
> RTTs; same interrupt times on servers; same etc, etc
> for both protocols.
>
> scott




More information about the NANOG mailing list