Re: [squid-users] Cache dir in tmpfs partition

From: Amos Jeffries <squid3_at_treenet.co.nz>
Date: Fri, 07 Jan 2011 18:11:24 +1300

On 05/01/11 08:25, Leonardo Rodrigues wrote:
>
> i dont have a clue how squid would deal with 2 cache_dir's, one being
> bigger and the other being smaller but on a fastest storage ....
>
> i really dont have a clue what to expect on that scenario, i have never
> done anything similar to that.
>
> sorry but i cannot help you with this idea ...
>
> Em 02/01/11 20:56, David Touzeau escreveu:
>> Thanks Leonardo
>>
>> If i create 2 caches :
>>
>> One (first) with 3Gb tmpfs memory
>> second with 500G hard disk memory.

Squid uses 2-4KB of index for every object. A general rule-of-thumb
calculation is that this averages out to 10MB of RAM for every GB of
cache storage.
  Your 500GB disk cache will consume around 5GB of RAM for indexing
alone. Although if you permit small (under 256 KB) objects to be stored
there its more likely that you will hit the 2^31 object count limit as well.

>>
>> Do you think that squid will increase performances using the first one.
>> Or did it make no sense ?
>>
>

Speed of the storage area is ignored. Size of available area and limits
on min/max object stored there are the only considerations.

Squid uses cache_mem as a RAM cache disk with a speedy format which
allows parts of the headers to be updated on 30x replies. The one use
I've seen where ramdisk was useful was to emulate a COSS directory in
squid-3. Unless you can dedicate a lot of RAM to the tmpfs you may see
no difference between a RAM disk and cache_mem.

Amos

-- 
Please be using
   Current Stable Squid 2.7.STABLE9 or 3.1.10
   Beta testers wanted for 3.2.0.4
Received on Fri Jan 07 2011 - 05:11:31 MST

This archive was generated by hypermail 2.2.0 : Fri Jan 07 2011 - 12:00:02 MST