Re: [squid-users] eyeballed performance question

The quick and dirty answer to your question would be that I'd expect
disk access time to be cut in half, so you should see service time
for objects cached to disk reduced significantly. That is just a SWAG.

On 8/26/05, john allspaw <jallspaw@yahoo.com> wrote:
> for those with high volume reverse proxy (http accel) experiences....

I run high volume proxies, but not reverse. I use SCSI almost exclusively,
primarily for scalability. However, I don't use Linux so my experiences
may not translate to your question.

>They have two cache_dirs, each 10Gb, on two SATA disks.
> cache_mem is set to 2048

Is this SATA150? 7200RPM drives?
When you say "each 10Gb", is the remainder of each drive used for
other purposes, or left quiescent?

> I'm about to replace them with 6-disk 15K RPM SCSI machines,
> same memory footprint. What kind of performance increase can I expect,
> if I were to put down a cache_dir of 10GB on each of the six disks ?

When you say that you'd put a cache_dir of 10GB on each of the six disks,
does this mean you're planning to use just the first (fastest) 10GB of
each disk (to keep the seek and access times low)?

That's a neat theory, but I've never seen it used in practice.

> my question is more about what can I expect from going from
> 2disk SATA to 6disk SCSI.

I assume this would be Ultra-320 SCSI?

If you have the machine available to test, I'd recommend using
bonnie++ to get a baseline on the new and old hardware, this
will give a good indication of the relative read and write performance
for any one disk in each design.

SCSI deals well with multiple drives on the same channel, but you
should strongly consider multiple channels and/or multiple cards.

Kevin Kadow
Received on Fri Aug 26 2005 - 11:10:35 MDT