Re: 2.6.9-rc3-mm3: vm-thrashing-control-tuning

[Andrew Morton]

Hideo AOKI <aoki@xxxxxxxxxxxxxxxxx> wrote:
>
> Since I made the patch for 2.6.9-rc3, the patch caused trouble 
> to sysctl code in -mm tree.
> 
> Attached patch fixes this issue.

Thanks.

Have you been doing any performance measurements on the thrash-control
code?

I went back to my original notes from when the patch was first being tested and
I had:

mem=256m, without, ./qsbench -p 4 -m 96

./qsbench -p 4 -m 96  27.50s user 3.92s system 9% cpu 5:34.26 total
./qsbench -p 4 -m 96  27.77s user 4.19s system 9% cpu 5:41.38 total
./qsbench -p 4 -m 96  27.22s user 4.17s system 9% cpu 5:16.75 total

with:

./qsbench -p 4 -m 96  27.40s user 2.08s system 35% cpu 1:23.67 total
./qsbench -p 4 -m 96  27.70s user 2.14s system 30% cpu 1:38.92 total
./qsbench -p 4 -m 96  27.03s user 1.79s system 39% cpu 1:13.16 total


But now I am unable to get anything remotely near the 1-2 minute runtimes
with this workload on current kernels.  Which means that either we broke it
again or I originally mismeasured it somehow.

I'm wondering if you've been able to notice any performance improvements
from the thrashing control and if so, how much and on what workload?
-
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/