On Saturday 01 March 2003 05:55 pm, Allan wrote:
> Hello,
>
> we are currently using a single server (squid 2.4-STABLE7, Linux
> RedHat 7.3, 1.4 GHz Pentium III, 2 Gb Ram, 4096 FileDescriptors) as
> reverse-proxy for a small site (approx 2,5K objects, about 150 Mb).
To extract the most out of this box, try
* Squid ufs over Linux tmpfs or
the null fs and large cache_mem & maximum_object_size_in_memory
* If this uses an Intel NIC, replace it with a 3com
* A kernel update -- the 2.4 jam patches are very nice.
> Due to growing load - 400 hits/sec, we are experincing loads about
> 30% user and 65% system-time
>
> Is this alarming? Should we consider bying "web"-switches and adding
> furter servers?
The system time seems a bit high for a squid server that should be 100%
in memory. The NIC and the filesystem are the likely culprits. At 400
req/sec, I would consider a trio of servers just for smoothing over
uprades or other downtime. Obviously the budget is not always so
flexible.
> As the site is quite small, would it be a good idea to build a
> "mini"-linux booting of floppy/CD with no harddrives/filesystems?
Not really necessary -- with that much RAM, yous shouldn't be touching
the hard drive, anyway. I would go with null storage or tmpfs for the
squid cache, though.
-- Brian
