Re: 2.6.4-mm2

From: Nick Piggin
Date: Fri Mar 19 2004 - 22:59:16 EST




Mark Wong wrote:

On Fri, Mar 19, 2004 at 06:50:26PM -0800, Andrew Morton wrote:

Mark Wong <markw@xxxxxxxx> wrote:

On Thu, Mar 18, 2004 at 07:41:50PM -0800, Andrew Morton wrote:

Andrew Morton <akpm@xxxxxxxx> wrote:

Mark, if it's OK I'll run up some kernels for you to test.

At

http://www.zip.com.au/~akpm/linux/patches/markw/

Ok, looks like I take the first hit with the 02 patch. Here's re-summary:

kernel 16 kb 32 kb 64 kb 128 kb 256 kb 512 kb
2.6.3 2308 2335 2348 2334
2.6.4-mm2 2028 2048 2074 2096 2082 2078
2.6.5-rc1-01 2394
2.6.5-rc1-02 2117
2.6.5-rc1-mm2 2036

Thanks, so it's the CPU scheduler changes. Is that machine hyperthreaded? And do you have CONFIG_X86_HT enabled?


Yes and CONFIG_X86_HT is enabled but I have hyperthreading disabled with
'acpi=off noht' (whichever one does it.)



The oprofile for the 01 kernel says
CPU: P4 / Xeon, speed 1497.76 MHz (estimated)
while the 02 kernel says
CPU: P4 / Xeon with 2 hyper-threads, speed 1497.57 MHz (estimated)
What's going on there?

Other than that, nothing in the kernel profile jumps out at me:
schedule, __copy_from_user_ll and __copy_to_user_ll are all
significantly lower *after* the CPU scheduler changes, which
is an indicator that cache behaviour is better.

Sar says average context switches/second were 9064 and 6567 before
and after.

The only thing I can see is the CPU utilisation averages show the
scheduler patches have more of a tendancy to load up one CPU more
before moving to another. This actually should be good behaviour,
generally but I wonder if it is hurting at all. I would be really
surprised if it was that significant.

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/