Re: [PATCH RFC] select_idle_sibling experiments

[Chris Mason <clm@fb.com>]

On Wed, Apr 13, 2016 at 05:40:20AM +0200, Mike Galbraith wrote:
> On Tue, 2016-04-12 at 22:45 +0100, Matt Fleming wrote:
> > On Sat, 09 Apr, at 01:30:34PM, Chris Mason wrote:
> > > 
> > > [ nohz throttling patch ]
> > > 
> > > I tested the nohz throttle two different ways, first with schbench's
> > > pipe simulation, it's easily 8% faster with messages bouncing between
> > > cpus.
> > > 
> > > In production it's hard to pick a single number because the benchmarks
> > > produce latency curves as the workload scales up in RPS.  The benefits
> > > range from 2-9% depending on the metric.  It's a nice win, and I'd love to
> > > see it go in.
> > 
> > Do we have any idea what the tradeoff is against power consumption for
> > throttling nohz?
> 
> That's measurable with the built in super duper watt meter gizmo
> (turbostat).  It should be dinky but existent, could be given an off
> button for particularly attentive laptop drivers to poke.  Servers
> drivers are unlikely to care given the performance win.

Our power sensors show its basically a wash during the production
benchmark runs.  Which makes sense because its really only blinking
on/off at very high frequency.

-chris