Re: [squid-users] cache sizes v. spindles

On Tue, 9 Mar 2004 jeff.richards@centrelink.gov.au wrote:

> I'm looking at moving our Squid servers, where cache is currently
> configured as 4 x 9G on 4 separate disks, to new hardware with 2 x 18G
> disks set aside for cache.
>
> Is it preferable, from a performance perspective, to configure cache on the
> new platform as 2 x 18G areas or 4 x 9G? Or no difference?

More spindles is almost always preferrable, but it also depends on the
quality of the spindles..

Squid eats seeek time. For this reason you want the sum(drive_seek_time /
number_of_drives) / number_of_drives to be as small.

Lets assume one configuration has 4 drives with 5 ms seek time, on the
other 2 drives with 4 ms seek time. Their relative "slowness" (smaller is
better) is then

4 x 5ms: ((5 / 4) * 4) / 4 = 5 / 4 = 1.2

2 x 4ms: ((4 / 2) * 2) / 2 = 4 / 2 = 2

Regards
Henrik
Received on Tue Mar 09 2004 - 01:36:35 MST