[lacnog] [v6ops] Windows CLAT Enters Private Preview: A Milestone for IPv6 Adoption | Microsoft Community Hub

Fernando Gont fgont en si6networks.com
Dom Nov 23 21:03:16 -03 2025


Hola, Ale,

Entenderia que el 20%-40% era con un cliente pegandole a un servidor 
dual stack. -- por otro lado, si el DC duera v6only y por acceder via v4 
ubiera una traduccion extra, entonces... no deberian haber muchas 
sorpresas...

Slds,
Fer




On 23/11/2025 12:51, Alejandro Acosta wrote:
> Hola Fer,
> 
>    Si, pero recuerda también puede haber NAT64/NAT46 en el otro extremo.
> 
> 
> Ale,
> 
> 
> On 23/11/25 11:45 AM, Fernando Gont wrote:
>> Hola, Henri,
>>
>> Hay otra cuestion: Para asegurarse de la causa, el delay extra deberia 
>> medirse en los primeros hops (donde se supone que existe la diferencia 
>> causada por el NAT). SI lo medis e2e, hay pueden existir otra cantidad 
>> de factores.
>>
>> Slds,
>> Fernando
>>
>>
>>
>>
>> On 22/11/25 11:01, Henri Alves de Godoy vía LACNOG wrote:
>>>
>>> Para quem quiser testar o realizar os testes em suas residências, o 
>>> arquivo curl-format.txt deve ter o conteúdo:
>>>
>>> time_namelookup:  %{time_namelookup}
>>> time_connect:     %{time_connect}
>>> time_appconnect:  %{time_appconnect}
>>> time_pretransfer: %{time_pretransfer}
>>> time_starttransfer:%{time_starttransfer}
>>> time_total:       %{time_total}
>>> size_download:    %{size_download}
>>> speed_download:   %{speed_download}
>>>
>>> Att,
>>> Henri.
>>>
>>> Em sáb., 22 de nov. de 2025 às 10:15, jordi.palet--- vía LACNOG 
>>> <lacnog en lacnic.net <mailto:lacnog en lacnic.net>> escreveu:
>>>
>>>     Gracias Henri,
>>>
>>>     Solo me queda la duda de si la diferencia puede ser por la WiFi. Yo
>>>     nunca hago pruebas de este tipo con Wifi, siempre con Ethernet,
>>>     porque puede haber variaciones/interferencias en la radio incluso
>>>     con pocos segundos de diferencia.
>>>
>>>     Eso si, discrepo contigo en que los usuarios no lo notan … te
>>>     aseguro que si, esa es la pelea de los operadores todos los días,
>>>     especialmente con los gammers!
>>>
>>>     Saludos,
>>>     Jordi
>>>
>>>     @jordipalet
>>>
>>>
>>>>     El 22 nov 2025, a las 14:00, Henri Alves de Godoy
>>>>     <henri.godoy en fca.unicamp.br <mailto:henri.godoy en fca.unicamp.br>>
>>>>     escribió:
>>>>
>>>>     Bom dia Jordi, Fernando e Alejandro,
>>>>
>>>>     Aproveitando essa manhã de Sábado ensolarado, realizei um pequeno
>>>>     teste básico a partir de minha residência, link Internet 600Mbps -
>>>>     Operadora Claro, Notebook Windows 11, utilizando Wi-Fi,
>>>>     dual-stack, IPv4 CGNAT.  IPv6 global.
>>>>
>>>>     Como destino está o website da Faculdade de Ciências Aplicadas -
>>>>     UNICAMP - Dual Stack - Apache Server.
>>>>
>>>>     Comando:
>>>>
>>>>     curl -4 -L --resolve www.fca.unicamp.br:443:143.106.230.5 `
>>>>       -o NUL -s -w "@curl-format.txt"
>>>>     https://www.fca.unicamp.br/test/10m.bin
>>>>     <https://www.fca.unicamp.br/test/10m.bin>
>>>>
>>>>     curl -6 -L --resolve www.fca.unicamp.br:443:2801:8a:c040:fca0::5 `
>>>>       -o NUL -s -w "@curl-format.txt"
>>>>     https://www.fca.unicamp.br/test/10m.bin
>>>>     <https://www.fca.unicamp.br/test/10m.bin>
>>>>
>>>>     Parâmetros analisados:
>>>>
>>>>     |time_connect|: handshake TCP
>>>>
>>>>      *
>>>>
>>>>         |time_appconnect|: handshake TLS (só em HTTPS)
>>>>
>>>>      *
>>>>
>>>>         |time_starttransfer| (TTFB): tempo até o primeiro byte
>>>>
>>>>      *
>>>>
>>>>         |time_total|: tempo total do download
>>>>
>>>>      *
>>>>
>>>>         |speed_download|: throughput final
>>>>
>>>>      *
>>>>
>>>>         Resultados:
>>>>
>>>>      *
>>>>
>>>>         IPv4
>>>>
>>>>      *
>>>>
>>>>         time_connect:     0.032 s
>>>>         time_appconnect:  0.643 s
>>>>         time_starttransfer:0.675 s
>>>>         time_total:       5.923 s
>>>>         speed_download:   1.77 MB/s
>>>>
>>>>      *
>>>>
>>>>         IPv6
>>>>
>>>>      *
>>>>
>>>>         time_connect:     0.043 s  (quase igual ao IPv4)
>>>>         time_appconnect:  0.101 s  (mais rapido)
>>>>         time_starttransfer:0.155 s  (mais rápido)
>>>>         time_total:       2.14 s   (mais rápido)
>>>>         speed_download:   4.89 MB/s
>>>>
>>>>      *
>>>>
>>>>
>>>>      *
>>>>
>>>>         Analise:
>>>>
>>>>      *
>>>>
>>>>
>>>>             TLS handshake (time_appconnect):
>>>>
>>>>
>>>>         IPv4 é 6,3 vezes mais lento no handshake
>>>>
>>>>      *
>>>>
>>>>
>>>>             IPv6: 0.101 s
>>>>
>>>>      *
>>>>
>>>>
>>>>             IPv4: 0.643 s
>>>>
>>>>      *
>>>>
>>>>      *
>>>>         Tempo total (time_total):
>>>>      *
>>>>
>>>>      *
>>>>         IPv4: 5.92 s
>>>>      *
>>>>
>>>>      *
>>>>         IPv6: 2.14 s
>>>>      *
>>>>
>>>>      *
>>>>
>>>>
>>>>           Throughput final
>>>>
>>>>         IPv6 está entregando 2,8 vezes mais throughput.
>>>>
>>>>      *
>>>>
>>>>
>>>>           IPv6: 4.89 MB/s
>>>>
>>>>      *
>>>>
>>>>
>>>>           IPv4: 1.77 MB/s
>>>>
>>>>      *
>>>>
>>>>      *
>>>>         É claro que é um teste único e seria interessante realizar uma
>>>>         média de várias conexões.
>>>>      *
>>>>
>>>>      *
>>>>         Os usuários conseguem então ter um experiência melhor de
>>>>         navegação ao consumir o website da Faculdade utilizando IPv6
>>>>      *
>>>>
>>>>      *
>>>>         Para um usuário normal e leigo, isso pode passar despercebido,
>>>>         pois o objetivo é acessar o website. Para nós profissionais,
>>>>         temos a certeza de que estamos proporcionando um acesso com
>>>>         uma experiência melhor para os nossos alunos e visitantes.
>>>>      *
>>>>
>>>>      *
>>>>         Abraços !
>>>>         Henri.
>>>>      *
>>>>
>>>>
>>>>     Em sáb., 22 de nov. de 2025 às 07:07, jordi.palet--- vía LACNOG
>>>>     <lacnog en lacnic.net <mailto:lacnog en lacnic.net>> escreveu:
>>>>
>>>>         Hola Henri, Fernando,
>>>>
>>>>         Creo que hablamos de diferentes mediciones … se trata de
>>>>         comparar IPv6 nativo extremo a extremo (por ejemplo
>>>>         residencial con IPv6-only y Facebook) con usuario residencial
>>>>         que, en el mismo operador y con los mismos dispositivo (para
>>>>         que sea comparable), tenga solo IPv4 y accede al mismo 
>>>> contenido.
>>>>
>>>>         He encontrado un paper, que quizás “contente" a Fernando,
>>>>         seguramente hay muchos otros (yo recuerdo haber leído otros de
>>>>         Akamai, Facebook y T-Mobile, pero ahora mismo no doy con ellos:
>>>>
>>>>         https://www.moritzsteiner.de/papers/Mobicom_IPv6.pdf
>>>> <https://www.moritzsteiner.de/papers/Mobicom_IPv6.pdf>
>>>>
>>>>         Lo he leído solo por encima, y no me queda claro si son las
>>>>         mismas pruebas que yo he hecho, parece que al menos son
>>>>         cercanas y comparables, solo que en red mobil. En este caso
>>>>         compara casos de diferentes operadores y muestra diferencias
>>>>         importantes entre ellos (por eso yo hago estas pruebas en mis
>>>>         despliegues, para asegurar que es óptimo). Por ejemplo en el
>>>>         caso de Verizon (el mejor de todos), incluso supera ese 40%
>>>>         que yo indicaba y copio textualmente:
>>>>
>>>>         In Figure 7(b), we observe similar reductions in PLTs for
>>>>         pages loaded by Verizon’s IPv6 clients over Verizon’s IPv6
>>>>         network. Specifically, we show that the median and 80% of the
>>>>         PLTs by IPv6 clients over Verizon’s IPv6 network are 48% and
>>>>         64% faster than PLTs over its IPv4 network, because of the
>>>>         significant differences in RTTs between Verizon’s IPv6 and
>>>>         IPv4 networks as shown in Figure 4(b).
>>>>
>>>>         Nota: PLT es Page Load Time, que creo que es equivalente a lo
>>>>         que yo indicaba de “time to complete http get”.
>>>>
>>>>         Hay otras partes del documento que por ejemplo indican mejoras
>>>>         en T-Mobile (red "pura” IPv6-only) del RTT entre el 49 y el 
>>>> 64%.
>>>>
>>>>         El documento tiene muchas otras referencias, quizás algunas
>>>>         sean de mediciones similares.
>>>>
>>>>         Saludos,
>>>>         Jordi
>>>>
>>>>         @jordipalet
>>>>
>>>>
>>>>>         El 21 nov 2025, a las 22:53, Henri Alves de Godoy vía LACNOG
>>>>>         <lacnog en lacnic.net <mailto:lacnog en lacnic.net>> escribió:
>>>>>
>>>>>         Hola Fernando y Jordi,
>>>>>
>>>>>         En LACNIC 34 ,  realizei um teste simples, realizando as
>>>>>         traduções via 464XLAT utilizando o Jool, apresentaram
>>>>>         pouquíssima variação de desempenho. Talvez isso ocorra porque
>>>>>         o mecanismo de transição opera diretamente no kernel Linux,
>>>>>         garantindo latências muito próximas entre IPv4 e IPv6, mesmo
>>>>>         sob carga.
>>>>>
>>>>>         Por curiosidade, realizarei outros em breve para comparar os
>>>>>         cenários atualmente.
>>>>>
>>>>>         Segue anexo.
>>>>>
>>>>>         Abraços !
>>>>>         Henri.
>>>>>
>>>>>         Em sex., 21 de nov. de 2025 às 17:09, Fernando Gont
>>>>>         <fgont en si6networks.com <mailto:fgont en si6networks.com>> 
>>>>> escreveu:
>>>>>
>>>>>             Hola, Jordi,
>>>>>
>>>>>             On 21/11/2025 15:52, jordi.palet--- vía LACNOG wrote:
>>>>>             > Hola Fernando,
>>>>>             >
>>>>>             > A mi FB nunca me dijo que tuviera dudas, y ademas
>>>>>             Akamai y otros
>>>>>             > hicieron las mismas mediciones y coincidían, y yo
>>>>>             también lo he
>>>>>             > comprobado en despliegues de clientes.
>>>>>
>>>>>             Repito: Cuales son las referencias a las publicaciones?
>>>>>
>>>>>
>>>>>
>>>>>             > Digo yo que tanta coincidencia,
>>>>>             > por algo será, además de la lógica de tener o no
>>>>>             multiples traducciones
>>>>>             > que sabemos que son latencia adicional.
>>>>>
>>>>>             Estamos hablando en serio?
>>>>>
>>>>>             Se supone que si uno va a tirar datos, se tiene que
>>>>>             entender, al menos
>>>>>             por arriba, las condiciones y limitaciones del
>>>>>             experimento, y en
>>>>>             principio un analisis (con sustancia) que permita
>>>>>             entender los
>>>>>             resultados obtenidos.
>>>>>
>>>>>             Respecto de las traducciones, alguien cuantifico la
>>>>>             latencia, de forma
>>>>>             mas o menos seria?
>>>>>
>>>>>             --             Fernando Gont
>>>>>             SI6 Networks
>>>>>             e-mail: fgont en si6networks.com 
>>>>> <mailto:fgont en si6networks.com>
>>>>>             PGP Fingerprint: F242 FF0E A804 AF81 EB10 2F07 7CA1 321D
>>>>>             663B B494
>>>>>
>>>>>             _______________________________________________
>>>>>             LACNOG mailing list
>>>>>             LACNOG en lacnic.net <mailto:LACNOG en lacnic.net>
>>>>>             https://mail.lacnic.net/mailman/listinfo/lacnog
>>>>> <https://mail.lacnic.net/mailman/listinfo/lacnog>
>>>>>             Cancelar suscripcion:
>>>>>             https://mail.lacnic.net/mailman/options/lacnog
>>>>> <https://mail.lacnic.net/mailman/options/lacnog>
>>>>>
>>>>>
>>>>>
>>>>>         --
>>>>> <Screenshot_2.png>_______________________________________________
>>>>>         LACNOG mailing list
>>>>>         LACNOG en lacnic.net <mailto:LACNOG en lacnic.net>
>>>>>         https://mail.lacnic.net/mailman/listinfo/lacnog
>>>>> <https://mail.lacnic.net/mailman/listinfo/lacnog>
>>>>>         Cancelar suscripcion:
>>>>>         https://mail.lacnic.net/mailman/options/lacnog
>>>>> <https://mail.lacnic.net/mailman/options/lacnog>
>>>>
>>>>
>>>>         **********************************************
>>>>         IPv4 is over
>>>>         Are you ready for the new Internet ?
>>>>         http://www.theipv6company.com <http://www.theipv6company.com/>
>>>>         The IPv6 Company
>>>>
>>>>         This electronic message contains information which may be
>>>>         privileged or confidential. The information is intended to be
>>>>         for the exclusive use of the individual(s) named above and
>>>>         further non-explicilty authorized disclosure, copying,
>>>>         distribution or use of the contents of this information, even
>>>>         if partially, including attached files, is strictly prohibited
>>>>         and will be considered a criminal offense. If you are not the
>>>>         intended recipient be aware that any disclosure, copying,
>>>>         distribution or use of the contents of this information, even
>>>>         if partially, including attached files, is strictly
>>>>         prohibited, will be considered a criminal offense, so you must
>>>>         reply to the original sender to inform about this
>>>>         communication and delete it.
>>>>
>>>>         _______________________________________________
>>>>         LACNOG mailing list
>>>>         LACNOG en lacnic.net <mailto:LACNOG en lacnic.net>
>>>>         https://mail.lacnic.net/mailman/listinfo/lacnog
>>>> <https://mail.lacnic.net/mailman/listinfo/lacnog>
>>>>         Cancelar suscripcion:
>>>>         https://mail.lacnic.net/mailman/options/lacnog
>>>>         <https://mail.lacnic.net/mailman/options/lacnog>
>>>>
>>>>
>>>>
>>>>     --
>>>
>>>
>>>     **********************************************
>>>     IPv4 is over
>>>     Are you ready for the new Internet ?
>>>     http://www.theipv6company.com <http://www.theipv6company.com>
>>>     The IPv6 Company
>>>
>>>     This electronic message contains information which may be privileged
>>>     or confidential. The information is intended to be for the exclusive
>>>     use of the individual(s) named above and further non-explicilty
>>>     authorized disclosure, copying, distribution or use of the contents
>>>     of this information, even if partially, including attached files, is
>>>     strictly prohibited and will be considered a criminal offense. If
>>>     you are not the intended recipient be aware that any disclosure,
>>>     copying, distribution or use of the contents of this information,
>>>     even if partially, including attached files, is strictly prohibited,
>>>     will be considered a criminal offense, so you must reply to the
>>>     original sender to inform about this communication and delete it.
>>>
>>>     _______________________________________________
>>>     LACNOG mailing list
>>>     LACNOG en lacnic.net <mailto:LACNOG en lacnic.net>
>>>     https://mail.lacnic.net/mailman/listinfo/lacnog
>>>     <https://mail.lacnic.net/mailman/listinfo/lacnog>
>>>     Cancelar suscripcion: https://mail.lacnic.net/mailman/options/lacnog
>>>     <https://mail.lacnic.net/mailman/options/lacnog>
>>>
>>>
>>>
>>> -- 
>>>
>>>
>>> _______________________________________________
>>> LACNOG mailing list
>>> LACNOG en lacnic.net
>>> https://mail.lacnic.net/mailman/listinfo/lacnog
>>> Cancelar suscripcion: https://mail.lacnic.net/mailman/options/lacnog
>>
> _______________________________________________
> LACNOG mailing list
> LACNOG en lacnic.net
> https://mail.lacnic.net/mailman/listinfo/lacnog
> Cancelar suscripcion: https://mail.lacnic.net/mailman/options/lacnog

-- 
Fernando Gont
SI6 Networks
e-mail: fgont en si6networks.com
PGP Fingerprint: F242 FF0E A804 AF81 EB10 2F07 7CA1 321D 663B B494



Más información sobre la lista de distribución LACNOG