Sluggish machine under 2.6.14 and 2.6.15-rc5

Sluggish machine under 2.6.14 and 2.6.15-rc5

[Richard Purdie]

I've recently acquired some new hardware which I'd image to be fairly
standard - 3.0Ghz P4 (630?) HT CPU (800Mhz FSB),  Gigabyte 8I945P-G
motherboard, 1GB memory. I installed ubuntu and their 2.6.12-smp based
kernel was fine but didn't support all my hardware (this board has some
ip8212 ide ports). I therefore installed 2.6.15-rc5 which ran badly and
made the machine slow and sluggish. 2.6.14 shows the same speed issue.
2.6.13 runs just as well as 2.6.12 does. I'm using smp kernels to try
and take advantage of hyperthreading.

By slow and sluggish, I mean that the machine is generally lacking in
processing power and is also unresponsive to user interaction. Under
2.6.12/2.6.13 compared to 2.6.14/15-rc5:

* kernel compiles are at least twice as fast
* desktop menus are much more responsive
* bringing up dialogs e.g. to "Log Off" happens much faster 
* machine boots up much faster

The .configs and config differences:
http://www.rpsys.net/openzaurus/temp/desktop/config-2.6.13
http://www.rpsys.net/openzaurus/temp/desktop/config-2.6.14
http://www.rpsys.net/openzaurus/temp/desktop/config.diff

dmesg and differences:
http://www.rpsys.net/openzaurus/temp/desktop/dmesg-2.6.13
http://www.rpsys.net/openzaurus/temp/desktop/dmesg-2.6.14
http://www.rpsys.net/openzaurus/temp/desktop/dmesg.diff

dmesg.diff is probably the most interesting.

Is there a known reason I'd see so much speed difference between these
kernels? Is there something simple I'm missing here? 

Sadly, the later kernels are basically unusable.

I'd like to avoid a binary search but if that's the next step...

Richard

Re: Sluggish machine under 2.6.14 and 2.6.15-rc5

[Richard Purdie]

On Wed, 2005-12-07 at 18:06 +0000, Richard Purdie wrote:
> ip8212 ide ports). I therefore installed 2.6.15-rc5 which ran badly and
> made the machine slow and sluggish. 2.6.14 shows the same speed issue.
> 2.6.13 runs just as well as 2.6.12 does. 

The issue was configuration. Somehow my use of make oldconfig had set a
number of debug options which started out unset. The spinlock changes of
2.6.13-git11 must have changed the overhead of spinlock debugging which
was the cause of the the slowdown I saw when comparing versions.

2.6.15-rc5 runs fine now.

Richard