Re: More performance tests of different AES implementations

From: Mok-Kong Shen (mok-kong.shen_at_t-online.de)
Date: 07/05/03 

Date: Sat, 05 Jul 2003 19:16:25 +0200

Brian Gladman wrote:
>
> ScottD wrote:
[snip]
> > 2) The statistical treatment of results is impressive, but making a few
> runs
> > and keeping the best result gives the most accurate and repeatable
> numbers.
>
> Are you saying that the timing results vary widely from one run to the next
> on your system(s)? If so this needs looking at.
[snip]

One point that I earlier remarked about Devine's first
benchmark was that the number of iterations was too
small. If the total running time is only of the order
of, say, 1 sec, I could hardly conceive that one could
obtain so many significant figures in the results as
a superficial look at the results (of the first
benchmark) would suggest. Because of the potential
inaccuracies problem, not only numerical but also due
to the varying operating conditions of the benchmark
run (I don't think that's absolutely constant even
without concurrent user applications), I use to employ
a very large number of iterations in my timing
measurement (it is based on the function clock, I
don't know how good it is), e.g. 10^9, in order to
even out the fluctuations of the results as far as
possible.

M. K. Shen

Relevant Pages

  • Re: Slow -f test on Windows in recent Perl-versions
    ... I upgraded from Perl version 5.005_03 to version v5.8.7 (an ActiveState ... Benchmark: timing 100 iterations of filetest... ...
    (comp.lang.perl.moderated)
  • Re: How to find regex at specific location on line
    ... benchmark complained about too few iterations. ... use Benchmark qw(:hireswallclock cmpthese timethese); ... timing 10000000 iterations of plain_regex, plain_substr, ...
    (perl.beginners)
  • Re: Replacement Help
    ... Mark Clements wrote: ... Benchmark: timing 100000 iterations of Clements... ...
    (comp.lang.perl.misc)
  • Re: Replacement Help
    ... Benchmark: timing 100000 iterations of PurlGurl... ... Prev by Date: ...
    (comp.lang.perl.misc)
  • Re: Replacement Help
    ... > Mark Clements wrote: ... > Benchmark: timing 100000 iterations of Clements... ...
    (comp.lang.perl.misc)