[Zlib-devel] Some more performance results

[Paul Marquess]

From: Zlib-devel-admin at zlib.net [mailto:Zlib-devel-admin at zlib.net]On

> On Friday, January 3, 2003, at 11:11  AM, Paul Marquess wrote:
> > 1.1.4				5.83
> > beta5 POSTINC		4.87
> > beta5				4.76
> > beta3				4.67
> > beta2 			4.56
> > beta3 POSTINC		4.55
>
> Paul,
>
> What is the uncertainty / repeatability of your measurements?  I'm
> wondering if I can compare these numbers at the tenth of a second
> level, or not.

I can repeat these figures to order - the numbers above are just a summary
of a test run that varied the input & output buffer sizes from 10k to 700k
in 10k increments. I've run the tests about half a dozen times, and the
relative order of each of the versions is always the same.

Each of my tests involves measuring the wall clock time taken to do 300
identical inflates in a tight loop, so the tenth of a second values are
probably correct.

The reliability of the figures is a good question. I'd be happier if I could
replicate the figures without Perl being part of the equation. If I can
steal a few cycles, I'll write a C equivalent of my test harness.

Paul