Re: [squid-users] Re: Squid naps each 3600 seconds !

From: Amos Jeffries <>
Date: Sun, 27 Oct 2013 19:28:22 +1300

On 26/10/2013 4:28 a.m., Omid Kosari wrote:
> Alex Rousskov wrote
>> On 10/24/2013 07:43 AM, Omid Kosari wrote:
>>> "digest_generation off" temporary solved problem but needs restart . I
>>> have
>>> tested with reload before .
>> Sounds like you have detected the source of the blocking Squid problem
>> and confirmed it! The fact that digest generation makes your Squid
>> unresponsive is essentially a Squid bug, but you might be able to help
>> Squid by adjusting its configuration. If you want to try it, I suggest
>> re-enabling digest generation and setting
>> digest_rebuild_chunk_percentage to 1.
>> If that works around your problem, great. Unfortunately, I suspect Squid
>> may still be unresponsive during digest generation time because of how
>> regeneration steps are scheduled and/or because of an old bug in the
>> low-level scheduling code of "heavy" events. However, using the minimum
>> digest_rebuild_chunk_percentage value (1) is still worth trying.
> Tried . Unfortunately problem persists .
> Alex Rousskov wrote
>>> But i could not use digest benefits anymore . Is there big penalty if
>>> both
>>> caches are in same gigabit switch ?
>> The digest optimization can be significant if the two Squids share a lot
>> of popular content _and_ it takes a long time for each Squid to get that
>> content from the internet during a miss.
>> HTH,
>> Alex.
> I mean if disable digest and just use HTCP .

That depends strngly on your traffic level and cache size.

HTCP uses bandwidth per-request, but no extra memory and no CPU
digesting activity.

Digest uses only periodic digest exchange bandwidth, but extra memory
and CPU generating the digests.

So if you have cache relatively smaller than traffic, digest works
better. Low traffic and huge cache size HTCP works better. It will be an
interesting study point to graph those and figure out what bandwidth and
cache sizes the swapover happens, I dont have anything beyond theory
about it though sorry.

Received on Sun Oct 27 2013 - 06:28:25 MDT

This archive was generated by hypermail 2.2.0 : Sun Oct 27 2013 - 12:00:06 MDT