Re: [PATCH] sched: high-res preemption tick

[Peter Zijlstra <a.p.zijlstra@chello.nl>]

On Sat, 2007-10-13 at 10:55 +0200, Peter Zijlstra wrote:
> On Sat, 2007-10-13 at 09:18 +0200, Mike Galbraith wrote:
> > On Fri, 2007-10-12 at 22:51 +0200, Peter Zijlstra wrote:
> > > Subject: sched: high-res preemption tick
> > > 
> > > Use HR-timers (when available) to deliver an accurate preemption tick.
> > 
> > This patch further reduced iperf context switching, and boosted
> > throughput.
> > 
> > iperf -c localhost -P 10 -t 300
> > 
> > Previously reported numbers
> > 
> > 2.6.23-smp
> > [SUM]  0.0-300.0 sec    153 GBytes  4.39 Gbits/sec
> > [SUM]  0.0-300.1 sec    148 GBytes  4.23 Gbits/sec
> > [SUM]  0.0-300.0 sec    152 GBytes  4.36 Gbits/sec
> > 
> > 2.6.23-smp-d (sched-devel)
> > [SUM]  0.0-300.0 sec    173 GBytes  4.96 Gbits/sec
> > [SUM]  0.0-300.1 sec    173 GBytes  4.96 Gbits/sec
> > [SUM]  0.0-300.0 sec    172 GBytes  4.93 Gbits/sec
> > 
> > Numbers from fresh pull today
> > 
> > 2.6.23-smp-d-hrt
> > (re-enable PREEMPT_RESTRICT)
> 
> Ah, but HRTICK is not compatible with PREEMPT_RESTRICT, it will be
> similar to !WAKEUP_PREEMPT.

(I do plan to fix that eventually, just need to do it)

Also, this seems to suggest iperf would like SCHED_BATCH.

> > [SUM]  0.0-300.1 sec    181 GBytes  5.19 Gbits/sec
> > [SUM]  0.0-300.0 sec    182 GBytes  5.22 Gbits/sec
> > [SUM]  0.0-300.1 sec    182 GBytes  5.22 Gbits/sec
> > 
> > 2.6.23-smp-d
> > [SUM]  0.0-300.1 sec    174 GBytes  4.97 Gbits/sec
> > [SUM]  0.0-300.1 sec    173 GBytes  4.95 Gbits/sec
> > [SUM]  0.0-300.1 sec    173 GBytes  4.96 Gbits/sec
> > 
> >