Re: [PATCH -v8][RFC] mutex: implement adaptive spinning

[Chris Mason <chris.mason@oracle.com>]

On Mon, 2009-01-12 at 18:24 +0100, Peter Zijlstra wrote:
> On Mon, 2009-01-12 at 12:14 -0500, Chris Mason wrote:
> > On Mon, 2009-01-12 at 17:50 +0100, Peter Zijlstra wrote:
> > > > 
> > > > (the file stat run is total run time, so lower is better.  The other
> > > > numbers are files or MB per second, so higher is better)
> > > > 
> > > > For the file create run, v8 had much lower system time than v7,
> > > > averaging 1s of sys time per proc instead of 1.6s.
> > > 
> > > Right, how about the spread in completion time, because that is the only
> > > reason I tried this fairness stuff, because you reported massive
> > > differences there.
> > > 
> > 
> > I reran the numbers with a slightly different kernel config and they
> > have changed somewhat.  These are just for the 4k file create run, all
> > numbers in files created per second (and the numbers are stable across
> > runs)
> > 
> > v8 avg 176.90 median 171.85 std 12.49 high 215.97 low 165.54
> > v7 avg 169.02 median 163.77 std 16.82 high 267.95 low 157.95
> 
> Any opinions on the fairness matter, will -v9 be unlocked and unfair
> again?

I'd rather have it simple than fair.  My benchmarks are pretty dumb, I
wouldn't want to add complexity just based on them.

-chris