[squid-users] Youtube not TCP_HIT Squid3.5.21-25

[Yuri]

Yup, thank you, Amos, for details. I do not remember all changelogs exactly.


29.05.2017 3:09, Amos Jeffries пишет:
> On 29/05/17 05:21, Yuri wrote:
>>
>> 28.05.2017 21:09, Eduardo Carneiro пишет:
>>> Ok. Tell me what technical details you need and I post here.
>> Configs. Logs. Topology. Usecases. Examples for single transactions.
>>> But if this were an ignoring cache-control issue, wouldn't that
>>> happen on
>>> squid 3.5.19 and previous versions as well?
>>>
>>> With the same conf works on 3.5.19 but not in newest squid3 versions.
>> Once more. Ignoring cache-control in squid is deprecated due to RFC
>> requirements. So, it leads to drops down with byte HIT in general. I do
>> not remember when it exactly desupported in squid's code, but seems
>> approx. at some 3.5.x.
>
> From 3.5.19 onwards ..
>
> 3.5.21 added support for Squid to differentiate between revalidation
> checks that were always vs sometimes necessary. Fixing the situation
> where responses had both a CC:private or CC:no-cache header and
> information stating cacheability "freshness".
>
> 3.5.22 fixed issues that were then seen with missing Last-Modified and
> old Date values, and the Collapsed Forwarding feature.
>
> 3.5.23 added support for caching objects with "Vary: *" which depended
> on the change in .21, and caching of 3034 responses. Also fixed the
> main outstanding issue behind Host verify MISS happening, and a
> regression in the .22 collapsed forwarding change.
>
> 3.5.24 fixed a regression in the "cache deny" access control handling
> which was causing some thing to be cached and revalidated when they
> should not have. This may make some things now un-cacheable if there
> is a misconfiguration in that directive.
>
> 3.5.25 fixed a stall that happens when SSL-Bump interacts with Host
> verification through the SNI value. While not strictly related, having
> it fixed allows a lot more Google/YouTube/Akamai transactions to reach
> the caching stages so their operations start to be visible instead of
> splice/tunnel'ed.
>
>
> The expected effect of these changes was to fix several major
> unnecessary bandwidth uses;
>  - cached YouTube videos starting mid-video when someone else using
> the proxy had only partially watched the same video - causing the user
> to force-refresh and purge the cached video.
>  - Chrome and similar Google originated downloads being a MISS for
> identical objects.
>  - the often cited terrible HIT reduction on some major hosting
> services (again Google, though also Akamai) traffic since 3.2 Host
> verify was added.
>  - collapsed forwarding resulting in large objects being stored to the
> cache then immediately deleted by a parallel fetch.
>
>
> Amos
> _______________________________________________
> squid-users mailing list
> squid-users at lists.squid-cache.org
> http://lists.squid-cache.org/listinfo/squid-users


-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 473 bytes
Desc: OpenPGP digital signature
URL: <http://lists.squid-cache.org/pipermail/squid-users/attachments/20170529/06f3e7d3/attachment.sig>