Re: performance....

On Tue, Apr 16, 2002 at 11:06:07AM -0500, Saurabh Desai wrote:
> 
> I used your benchmark program for this (for yield tests).
> 
> On Mon, Apr 15, 2002 at 06:32:14PM -0500, Saurabh Desai wrote:
> >
> > Actually, on my 4-way system, I am seeing exactly other way around.
> > NGPT - about 5 secs.
> > Linuxthreads - about 48 secs.
> 
> What benchmark are you using?  The same one I did?  Or
> something else?
> 
> --
> Richard Seaman, Jr.        email:    [email protected]
> 5182 N. Maple Lane         phone:    262-367-5450
> Nashotah WI 53058            fax:    262-367-5852

Well, I re-ran the tests here, which is UP, just to make sure
I didn't accidentally reverse the results.  I still get basically
the same result (I can't get access to a quiet system at the
moment, so the results are off marginally from what I posted 
since there is a small load on the system.

I'd be interested if you can confirm my general results on
a UP system?

If so, it would be very interesting to try to understand why
the results would differ so dramatically from a UP to a
4 way system.  It would also be interesting to know what
kernel and scheduler you're running.  If there is such a
dramatic difference between a UP and a 4-way box (factor of
6 one way vs factor of 8 or 9 the opposite way), I wonder
if it doesn't point up some problem in the kernel scheduler
for 4 way vs. UP?


-- 
Richard Seaman, Jr.        email:    [email protected]
5182 N. Maple Lane         phone:    262-367-5450
Nashotah WI 53058            fax:    262-367-5852