On Thu, Aug 10 2006, David Miller wrote:
> From: Jens Axboe <[email protected]>
> Date: Thu, 10 Aug 2006 13:06:27 +0200
>
> > run_timer_softirq+0x0/0x18e: took 3750
> > run_timer_softirq+0x0/0x18e: took 2595
> > run_timer_softirq+0x0/0x18e: took 6265
> > run_timer_softirq+0x0/0x18e: took 2608
> >
> > So from 2.6 to 6.2msecs just that handler, auch. During normal running,
> > the 2.6 msec variant triggers quite often.
>
> It would be interesting to know what timers ran when
> the overhead got this high.
>
> You can probably track this with a per-cpu array
> of pointers, have run_timer_softirq record the
> t->func pointers into the array as it runs the
> current slew of timers, then if the "took" is
> very large dump the array.
I did that, and it got me nowhere, no timers are to blame. Trying to
reproduce the slowdowns reported, these tests were with full preemtion
enabled. Running the exact same test that showed:
centera:/sys/block/sdc/queue # cat complete
8596, 8148, 6833, 7768, 11332, 7781, 8652, 6956, 6716, 13710
before (almost 14 msecs!!), now shows:
centera:/sys/block/sdc/queue # cat complete
172, 138, 125, 152, 122, 179, 157, 155, 142, 155
which is a hell of a lot more reasonable. Preempting the io completion
is, well, shall we say pretty suboptimal. So much for improving latency
:-)
--
Jens Axboe
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
[Index of Archives]
[Kernel Newbies]
[Netfilter]
[Bugtraq]
[Photo]
[Stuff]
[Gimp]
[Yosemite News]
[MIPS Linux]
[ARM Linux]
[Linux Security]
[Linux RAID]
[Video 4 Linux]
[Linux for the blind]
[Linux Resources]
