On Wed, 2005-12-21 at 17:37 -0500, Steven Rostedt wrote:
> On Wed, 2005-12-21 at 01:04 -0500, Lee Revell wrote:
> > Here is a 13ms+ trace caused by "netstat -anop". I noticed that the
> > preempt count is 0 so there are no locks to drop, would adding a
> > cond_resched() be acceptable? Or should this be considered more
> > evidence of the need for softirq preemption?
> >
> > preemption latency trace v1.1.5 on 2.6.14-rt22
> > --------------------------------------------------------------------
> > latency: 13824 us, #16533/16533, CPU#0 | (M:preempt VP:0, KP:0, SP:0 HP:0)
> > -----------------
> > | task: softirq-timer/0-3 (uid:0 nice:0 policy:1 rt_prio:1)
> > -----------------
> >
> > _------=> CPU#
> > / _-----=> irqs-off
> > | / _----=> need-resched
> > || / _---=> hardirq/softirq
> > ||| / _--=> preempt-depth
> > |||| /
> > ||||| delay
> > cmd pid ||||| time | caller
> > \ / ||||| \ | /
> > netstat-29157 0D.h2 0us : __trace_start_sched_wakeup (try_to_wake_up)
> > netstat-29157 0D.h2 1us : __trace_start_sched_wakeup <<...>-3> (62 0)
> > netstat-29157 0D.h. 2us : wake_up_process (wakeup_softirqd)
>
> Hi Lee, I've seen the same thing here. I wrote up this patch as a case
> study to see if it can work and it does. Now this sacrifices memory for
> speed. I added a bitmask in the inet_hash to keep track of all the
> buckets that are used, and use the find_next_bit to search the list in
> established_get_first.
>
> After running the netstat -anop I get the following:
>
> preemption latency trace v1.1.5 on 2.6.14-rt22
> --------------------------------------------------------------------
> latency: 377 us, #225/225, CPU#0 | (M:preempt VP:0, KP:0, SP:0 HP:0)
> -----------------
> | task: softirq-timer/0-3 (uid:0 nice:0 policy:1 rt_prio:1)
> -----------------
>
> _------=> CPU#
> / _-----=> irqs-off
> | / _----=> need-resched
> || / _---=> hardirq/softirq
> ||| / _--=> preempt-depth
> |||| /
> ||||| delay
> cmd pid ||||| time | caller
> \ / ||||| \ | /
> <idle>-0 0D.h4 0us : __trace_start_sched_wakeup (try_to_wake_up)
> <idle>-0 0D.h4 0us : __trace_start_sched_wakeup <<...>-3> (62 0)
> <idle>-0 0Dnh3 0us : try_to_wake_up <<...>-3> (62 8c)
> <idle>-0 0Dnh2 1us : preempt_schedule (try_to_wake_up)
> <idle>-0 0Dnh2 1us : wake_up_process (wakeup_softirqd)
> [snip]
>
> So it really does improve the latency here. Now is this worth the
> overhead? This might be useful in other places to.
Any chance you can regenerate the patch against 2.6.15-rc5-rt4?
Lee
-
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]