Re: [squid-users] High CPU usage for large object

The place to start is grabbing traces from a running system to find out
where the time is being spent.

You'll probably want to start with vmstat and see whether its chewing 100%
of one CPU, or whether its blocked waiting for disk IO, or a combination
of the two.

Its probably CPU - I'd then run oprofile to gather statistics.

If you've paid for AS then you've paid for support; please consider contacting
Redhat and asking them for assistance.

Adrian

On Tue, Aug 07, 2007, NGUYEN, KHANH, ATTSI wrote:
> Hi,
>
> I am using squid 2.6 on Linux AS version 4, update 3.
>
> Hardware: dell 2850, 4 GB memory, 6 x 72 GB disks. NO RAID. Each disk is one mount point.
>
> Squid basic configuration:
>
> cache_mem: 2 GB
> maximum_object_size 5096 MB
> maximum_object_size_in_memory 100 MB
> cache_replacement_policy lru
> 6 cache_dir on each disk.
> Cache serve is configured as a reverse proxy in front of an apache server.
>
> squid compilation option: enable-follow-x-forwarded-for, enable-async-io, enable-auth, disable-wccp, enable-snmp, enable-x-accelerator-vary, enable-remove-policies=lru
>
> When I request a 4GB object from the squid server(the object is already cached) vs from an apache server (version 2.2.0), the cpu usage of the squid process is at least 3 times more than the cpu usage of the apache server. This object size exceeds the maximum_object_size_in_memory thus it has to get from the disk each time there is a request for it. So perhaps the squid has some extra overhead. However, 3 times more seems unusual. Any body has any suggestion on tuning the squid or the OS to better serve large object? Also I notice that the cpu takes a lot of hit when the object is greater than 1 MB.
>
> Thanks,
> Khanh

-- 
- Xenion - http://www.xenion.com.au/ - VPS Hosting - Commercial Squid Support -
- $25/pm entry-level bandwidth-capped VPSes available in WA -