develooper Front page | perl.perl5.porters | Postings from June 2006

Re: speed issue (was: Its time we set the score straight on Perl5 and Perl 6 ...)

Thread Previous | Thread Next
Jim Cromie
June 17, 2006 13:56
Re: speed issue (was: Its time we set the score straight on Perl5 and Perl 6 ...)
Message ID:
Dave Mitchell wrote:
> On Fri, Jun 16, 2006 at 04:25:33PM +0200, demerphq wrote:
>> I think we need to take some time on the speed issue.
> Is there a speed issue?
yes - the issue is that its hard to tell whether the 'average' change 
helps or not.

In my limited experience, perlbench has 2-4 % noise in repeated tests on 
the same perl.
This will hide many real perf effects.  I had expected 'some' effect 
from the arena hacks
I submitted, but couldnt construct a test-case file that showed anything 
(w my expectations)

I vaguely recall even purposely pessimizing some stuff to see if it 
slowed down.
I didnt get any satisfaction.

Which is not to blame perlbench;  I looked a bit at it, and
(no suprise) I couldnt see anything obviously wrong.
It may be that a malloc-happy prog like perl will forever be stochastic
in its performance.

The last 5% is hard - in everything, and Im certainly too lazy to 
seriously contemplate
the idea that I can do better than Gisle.  In touching perlbench, there 
are 95 ways
backwards, and 5 forward.  Those are not good odds for a random walk.

Thread Previous | Thread Next Perl Programming lists via nntp and http.
Comments to Ask Bjørn Hansen at | Group listing | About