[sclug] Hard disk performance question

[lug at assursys.co.uk]

On Thu, 26 Jun 2003, Derek M Jones wrote:

> Neil,
> 
> >I have two IDE hard drives fitted:
> >
> >Seagate Barracuda 40Gb 7200rpm on Primary Master  (/dev/hda)
> >IBM Deskstar 60GXP 40Gb 7200rpm on Secondary Master (/dev/hdc)
> >
> >The IBM is about 50% slower than the Seagate. Any ideas if I can improve this by tweaking? Here are the parametrs and performance figures:
> 
> Performance is not just driven by rpm.

...but buffer/cache size, caching algorithm, firmware, data density, and
probably more besides. ;-)

A couple of tweaks that Neil hasn't done are;

- enable unmasking of IRQs - though that probably won't help much

- try experimenting with different multi-sector transfer sizes. Again, these
  are usually set optimally by the firmware.

- check if both discs are running at their fastest DMA settings. 'hdparm -i
  /dev/hdx' will show you the modes available, and the currently selected mode
  is marked with an asterisk.

> I found that putting the swap partition on the least often accessed drive
> made a bug difference.

That won't make any difference here, as Neil's benchmarking using the parent
block devices (i.e. /dev/hdx) which should be reading from the rim of the
disc - i.e. the fastest part (though these days, physical and logical
geometry may not necessarily tally!)

Personally, I have two discs, so I stripe my swap partition across the two,
RAID0-style (the swap layer does this automagically if you enable swap on
more than one device).

> derek

Best Regards,
Alex.
-- 
Alex Butcher      Brainbench MVP for Internet Security: www.brainbench.com
Bristol, UK                      Need reliable and secure network systems?
PGP/GnuPG ID:0x271fd950                         <http://www.assursys.com/>