Re: [PATCH [RT] 10/14] adjust pi_lock usage in wakeup

["Peter W. Morreale" <pmorreale@novell.com>]

On Thu, 2008-02-21 at 11:48 -0500, Steven Rostedt wrote:
> On Thu, 21 Feb 2008, Gregory Haskins wrote:
> 
> > From: Peter W.Morreale <pmorreale@novell.com>
> >
> > In wakeup_next_waiter(), we take the pi_lock, and then find out whether
> > we have another waiter to add to the pending owner.  We can reduce
> > contention on the pi_lock for the pending owner if we first obtain the
> > pointer to the next waiter outside of the pi_lock.
> >
> > This patch adds a measureable increase in throughput.
> 
> I see how this may decrease contention (slightly less time in holding the
> pi_lock). But, please, when stating something like: "adds a measurable
> increase in throughput", show the benchmark numbers.
> 
> -- Steve
> 

Approximately 3% to the dbench benchmark we used.

My "standard" sanity check was to mount a ramfs filesystem and execute:

dbench -t 10 30

five times and generate an average from the reported "Throughput"
numbers displayed by the runs. 

dbench was chosen merely because of the contention on dcache_lock.

Best,
-PWM