Re: requests/hr per spindle

On Wed, 11 Nov 1998 dinn@moose.to wrote:

> Has anyone got any empirical data as to how many requests/hr squid can
> sustain per average fast hard disk?

We have 4 squid boxes configured with equal weight load-balancing
on an Alteon switch, and the peak hourly average over a few days is
64,556 and 58,049 requests an hour using a 10,000rpm IBM cache disk and
a 7,200rpm Barracuda disk respectively. Otherwise, there's no difference
in hardware between them. The Alteon does load balancing based on the
number of simultaneous connections, so the quicker the box, the more
requests it gets.

> I'm approaching ~100K hits/hr with a pair of 9G wide 'cudas on a single
> controller and while I appear to have plenty of RAM free (~80M out of 256)
> things are starting to slow a little.

Yep, that appears to be right at the limit we've seen too.

> CPU utilization is <20%, typically, so it's not the fault of the P2/300.
>
> If it's not ram or CPU power, then it has to be disks. I'm thinking that the
> Cudas are at fault - I can't see it being the SCSI bus itself.

I once tried to find out whether the SCSI system made much of a
difference, and I was only a 5.7% improvement between using Ultra
Wide over old SCSI2 with the same disk, presumably because there's
less latency.

> Anyone else vote that it's time to double my RAM (thereby reducing load on
> the drives) or time to add another spindle?

We've found that providing there's enough RAM to cache the spool
directory hierachy, seek time is the number one factor for increased
number of connections.

You should be able to effectively increase the peak number of connections
by using Async-IO, but it doesn't work for me on FreeBSD just yet :-*(

Chris Foote SE Net
Technical Manager 222 Grote Street
SE Network Access Adelaide SA 5000
e-mail chris@senet.com.au Australia
phone : (08) 8221 5221 PGP Public Key available from
fax: (08) 8221 5220 http://www.senet.com.au/PGP
support: (08) 8221 5792
Received on Wed Nov 11 1998 - 19:18:53 MST