Re: [squid-users] Problem with TCP_MEM_HITs

From: Santiago del Castillo <santiago.del.castillo@dont-contact.us>
Date: Thu, 02 Nov 2006 10:54:11 -0300

Hi, sry, when I said maximum_object_size I wanted to say
maximum_object_size_in_memory :P.

Right now (i don't know how) it's working. Here is some info:

Cache information for squid:
        Request Hit Ratios: 5min: 92.0%, 60min: 92.3%
        Byte Hit Ratios: 5min: 35.5%, 60min: 34.7%
        Request Memory Hit Ratios: 5min: 61.6%, 60min: 60.2%
        Request Disk Hit Ratios: 5min: 0.5%, 60min: 0.4%

Memory usage for squid via mallinfo():
        Total space in arena: 79176 KB
        Ordinary blocks: 71934 KB 9224 blks
        Small blocks: 0 KB 0 blks
        Holding blocks: 432 KB 2 blks
        Free Small blocks: 0 KB
        Free Ordinary blocks: 7241 KB
        Total in use: 72366 KB 91%
        Total free: 7241 KB 9%
        Total size: 79608 KB

It's very VERY good compared with I was getting before.

My box is a P4 3.0 with 4 GB RAM under FedoraCore 5.

Cheers and thanks,
Santiago del Castillo

K Kadow wrote:
> On 11/1/06, Santiago del Castillo <santiago.del.castillo@cloncom.com>
> wrote:
>> Hi, i'm having problems with MEM_HITs, the mem_cache is 2GB, the
>> maximum_object_size is 4096 KB and i'm having just 10.7 % MEM_HIT ratio.
>
> What is your maximum_object_size_in_memory?
> The default is very low.
>
>
> I've been told here that, depending on your host OS and how much RAM
> you have, you can actually increase hit ratios and performance by
> setting both cache_mem and maximum_object_size_in_memory very low, and
> relying on the operating system to intelligently use memory to cache
> disk blocks and serve disk hit answers out of the OS cache.
>
> This assumes the server is dedicated to Squid and that the OS is
> smarter about caching frequently used objects in memory than Squid is.
>
> Kevin
>
Received on Thu Nov 02 2006 - 06:54:47 MST

This archive was generated by hypermail pre-2.1.9 : Fri Dec 01 2006 - 12:00:02 MST