Re: [squid-users] cache_mem

From: Adrian Chadd <adrian_at_squid-cache.org>
Date: Fri, 23 Jan 2009 02:05:28 -0500

2009/1/22 Amos Jeffries <squid3_at_treenet.co.nz>:

>> How intensive is intensive? At the moment squid is averaging a mere 2.4%
>> processor time.
>
> IIRC older Squid-2 had to step a linked-list the length of the object in 4KB
> chunks to perform one of the basic operations (network write I think).

Yeah - the memory cache in Squid-2 was really only initially designed
as a sort of data pipeline between the server, the store, and the
client-side. It sort of "grew" the stuff needed to be a memory cache
by virtue, iirc, of wanting to support one incoming stream -> multiple
client retrievals without having to always go via the disk store for
it.

Unless you need the extra boost it gives you in very specific circumstances:

* use low cache_mem; but if you notice that you're hitting the disk often;
* use a larger cache_mem; but keep maximum_object_size_in_memory down
to around 64k

Squid-3 "sort of" fixed this. It wasn't ever fully fixed, much like
how the problem could be fixed in Squid-2 if someone wanted to do the
slight trickery required.

Adrian
Received on Fri Jan 23 2009 - 07:05:33 MST

This archive was generated by hypermail 2.2.0 : Fri Jan 23 2009 - 12:00:02 MST