Re: [sched,rcu] 9234566d3a3: +1.6% will-it-scale.scalability, +1302.6% time.involuntary_context_switches

From: Paul E. McKenney
Date: Mon Apr 28 2014 - 12:37:43 EST


On Tue, Apr 22, 2014 at 10:41:12AM +0800, Jet Chen wrote:
> On 04/22/2014 09:59 AM, Paul E. McKenney wrote:
> >On Mon, Apr 21, 2014 at 02:28:21PM +0800, Jet Chen wrote:
> >>Hi Paul,
> >>
> >>we noticed the below changes on
> >>git://git.kernel.org/pub/scm/linux/kernel/git/paulmck/linux-rcu.git torture.2014.04.18a
> >>commit 9234566d3a36c0aead8852e3c2ca94cd8ebfe219 ("sched,rcu: Make cond_resched() report RCU quiescent states")
> >
> >My impression of these statistics is that this commit results in huge
> >numbers of additional context switches and interrupts, but has a slightly
> >positive effect on performance and a larger negative effect on
> >scalability.
> >
> >Is this a reasonable interpretation?
> >
> Yes, you're right.

OK, sounds like a bit of rework is in order, then...

Thanx, Paul

> >>Comparison 1 - parent commit of 9234566d3a36c0aead8852e3c2ca94cd8ebfe219 vs 9234566d3a36c0aead8852e3c2ca94cd8ebfe219
> >>
> >>e119454e74a852f 9234566d3a36c0aead8852e3c
> >>--------------- -------------------------
> >> 1035948 ~ 0% +1.6% 1052990 ~ 0% TOTAL will-it-scale.per_thread_ops
> >> 1271322 ~ 0% +1.8% 1294004 ~ 0% TOTAL will-it-scale.per_process_ops
> >> 0.63 ~ 0% -5.2% 0.60 ~ 0% TOTAL will-it-scale.scalability
> >> 22470 ~ 2% +1302.6% 315168 ~ 2% TOTAL time.involuntary_context_switches
> >> 84265 ~ 5% +1047.1% 966581 ~ 1% TOTAL interrupts.IWI
> >> 1828 ~44% +189.6% 5295 ~13% TOTAL time.voluntary_context_switches
> >> 5337 ~ 1% +82.1% 9720 ~ 1% TOTAL vmstat.system.cs
> >> 118599 ~ 0% -30.4% 82545 ~ 0% TOTAL interrupts.0:IO-APIC-edge.timer
> >> 224021 ~ 4% +34.7% 301858 ~ 2% TOTAL interrupts.RES
> >> 25148 ~ 0% +7.0% 26917 ~ 0% TOTAL vmstat.system.in
> >> 7063439 ~ 0% -5.2% 6694536 ~ 0% TOTAL interrupts.LOC
> >> 188866 ~ 0% -3.1% 183008 ~ 0% TOTAL interrupts.NMI
> >> 188866 ~ 0% -3.1% 183008 ~ 0% TOTAL interrupts.PMI
> >> 3720 ~ 0% -1.5% 3665 ~ 0% TOTAL time.system_time
> >> 1215 ~ 0% -1.4% 1198 ~ 0% TOTAL time.percent_of_cpu_this_job_got
> >>
> >>
> >>Comparison 2 - b84c4e08143c98dad4b4d139f08db0b98b0d3ec4 vs 9234566d3a36c0aead8852e3c2ca94cd8ebfe219
> >>
> >>Fengguang has reported stats changes about b84c4e08143c98dad4b4d139f08db0b98b0d3ec4 to you days ago.
> >>git://git.kernel.org/pub/scm/linux/kernel/git/paulmck/linux-rcu.git dev.2014.04.14a
> >>commit b84c4e08143c98dad4b4d139f08db0b98b0d3ec4 ("sched,rcu: Make cond_resched() report RCU quiescent states")
> >>
> >>Let's have a compare here.
> >>
> >>b84c4e08143c98d 9234566d3a36c0aead8852e3c
> >>--------------- -------------------------
> >> 809309 ~ 0% -2.6% 788400 ~ 0% TOTAL will-it-scale.per_process_ops
> >> 0.61 ~ 0% -1.9% 0.60 ~ 0% TOTAL will-it-scale.scalability
> >> 434080 ~ 0% -1.5% 427643 ~ 0% TOTAL will-it-scale.per_thread_ops
> >> 4 ~11% +1.2e+05% 5249 ~ 2% TOTAL interrupts.IWI
> >> 607 ~ 7% +28.0% 778 ~14% TOTAL interrupts.47:PCI-MSI-edge.eth0
> >> 12349 ~ 2% -14.6% 10548 ~ 1% TOTAL interrupts.0:IO-APIC-edge.timer
> >> 3078 ~ 3% +20.9% 3722 ~ 6% TOTAL interrupts.RES
> >>
> >>
> >>Comparison 3 - parent commit of b84c4e08143c98dad4b4d139f08db0b98b0d3ec4 vs b84c4e08143c98dad4b4d139f08db0b98b0d3ec4
> >>
> >>Duplicated with Fengguang's report. Help you to see these info in a page :)
> >>
> >>ad86a04266f9b49 b84c4e08143c98dad4b4d139f
> >>--------------- -------------------------
> >> 676264 ~ 0% +3.3% 698461 ~ 0% TOTAL will-it-scale.per_thread_ops
> >> 1174547 ~ 0% +3.0% 1209307 ~ 0% TOTAL will-it-scale.per_process_ops
> >> 1.67 ~ 0% -2.3% 1.63 ~ 0% TOTAL will-it-scale.scalability
> >> 10522 ~ 2% +921.2% 107463 ~ 1% TOTAL time.involuntary_context_switches
> >> 77671 ~ 3% +67.0% 129688 ~ 3% TOTAL interrupts.RES
> >> 99502 ~ 0% -27.8% 71813 ~ 0% TOTAL interrupts.0:IO-APIC-edge.timer
> >> 2554 ~ 0% +49.1% 3809 ~ 1% TOTAL vmstat.system.cs
> >> 11524 ~ 0% -2.3% 11259 ~ 0% TOTAL vmstat.system.in
> >> 213 ~ 0% -4.3% 204 ~ 0% TOTAL time.system_time
> >> 74 ~ 0% -4.1% 71 ~ 0% TOTAL time.percent_of_cpu_this_job_got
> >> 3495099 ~ 0% -3.1% 3387173 ~ 0% TOTAL interrupts.LOC
> >>
> >>
> >>
> >>
> >>Thanks,
> >>Jet
> >>
> >>
> >>
> >
> >>./runtest.py open2 32 1 4 6 8
> >>
> >>
> >
>

--
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/