Re: [squid-users] What load can SQUID cope with in accelerator mode?

On Wed, 14 Mar 2001, Simon Morley wrote:

> Hi,
> I am trying to work out how much load I can expect to handle using SQUID in
> accelerator mode. It would very much help if people could give me an idea of
> what load they are coping with for a given hardware configuration and OS.
>
> For example we have the following hardware:-
>
> Dual PIII 800Mhz with 1GB RAM and 8GB SCSI for the cache.
> This is running Linux with a 2.2 kernel.
>
> SQUID is in accelerator mode and is running with a pool of 30 redirectors
> which munge URL's from public to internal representation. This all front
> end's a pool of Windows 2000 servers which runs our service.
>
> At peak load we are getting about 100 requests a second, of which I am
> getting a 95% hit rate with the average object size being around 6KB. This
> uses about 6 to 7% CPU and around 2 or 3 of the redirectors to be in
> permanent use.
>
> This does suggest that we will run out of bandwidth before SQUID runs out of
> power, but I want to get more *real world* values.
>
> Any help would be very much appreciated.

I've done tests with polygraph and achieved 500+ requests per second
with similar workload characteristics (90% DHR, 10K file size).

One of the first bottlenecks that you'll probably hit is the
number of ephemeral TCP ports. In my case I couldn't get past
300 req/sec until I decreased the TCP MSL value so ports are
reused faster. On your linux box you'll want to increase
the ephemeral port range first, and decrease MSL as a last resort.
Received on Wed Mar 14 2001 - 11:27:22 MST