2.4 vs 2.6

[Cef]

On Mon, 13 Sep 2004 17:24, John wrote:
> 2.4 kernels outperform 2.6 at disk I/O on all systems I've tried them on.
> Knoppix 3.6  2.4 on the Ubuntu box 18.5
>              2.6                   15.5
>              2.4 on my SuSE box    30 SuSE manages 29.5
>              2.6                  <10

What are the figures measured in? Units please.

What tool/test was used to get these figures?

> U Intel BX
> S Intel 815

U for Ubuntu and S for SuSE I'm guessing?

> The problem I originally noticed it on is a Gigabyte mobo, SiS chipset.
> I think I reported the same sort of thing on my Acerpower 4400 (Via),
> but I don't care to reboot that atm to check it. Ditto my server (Also BX).

Errrgh! SiS. I've had enough bad stuff with SiS to last me a lifetime.

PS: hdparm's -t test is not a valid comparison between 2.4 and 2.6 as hdparm 
does a very vague test that cannot be considered reliable for anything other 
than basic testing. And if you're going to run hdparm as a test, please 
provide both figures from -tT and the output of -v with your tests. If you 
are looking for a better tool, look at bonnie++.

PPS: Try all your tests with 'elevator=deadline' added to the 2.6 kernel 
commandline and see if this improves your situation. Feel free to read all 
the files in Documentation/block/ of the kernel source to see how to tune the 
IO elevator/scheduler.

-- 
 Stuart Young - aka Cefiar - cef at optus.net