Is this ext2 versus ext3? If so, that's a pretty good result isn't it? I
forget what kernbench looked like for stock ext3.
> SDET 32 (see disclaimer)
> Throughput Std. Dev
> 2.5.66-mm2 100.0% 0.7%
> 2.5.66-mm2-ext3 4.7% 1.5%
Yes, this is presumably a lot more metadata-intensive, so we're just
hammering the journal semaphore to death. We're working on it.
> ftp://ftp.kernel.org/pub/linux/kernel/people/mbligh/benchmarks/2.5.66-mm2-ext3/
Offtopic, a raw sdet64 profile says:
5392317 total
4478683 default_idle
307163 __down
169770 .text.lock.sched
106769 schedule
88092 __wake_up
57280 .text.lock.transaction
I'm slightly surprised that the high context switch rate is showing up so
much contention in sched.c. I'm assuming that it's on the sleep/wakeup path
and not in the context switch path. It would be interesting to inline the
spinlock code and reprofile.
We really should be using the waker-removes-wakee facility in the semaphore
code, but that's not completely trivial.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/