develooper Front page | perl.perl5.porters | Postings from January 2001

Re: 2% speedup with Duff's device

Thread Previous | Thread Next
Nicholas Clark
January 23, 2001 06:49
Re: 2% speedup with Duff's device
Message ID:
On Tue, Jan 23, 2001 at 08:30:11AM -0600, Jarkko Hietaniemi wrote:
> On Tue, Jan 23, 2001 at 01:02:42PM +0000, Nick Ing-Simmons wrote:
> > Nicholas Clark <> writes:
> > >If one uses Duff's device in PERL_HASH, the regression tests go about 2%
> > >faster.
> > 
> > Note - that may be literally true, but is not the speed difference in perl.
> > The tests are riddled with sleep()s and other system timing overhead.
> That is true, the timing of the whole suite is to be taken with a
> biiig grain of salt -- but a test of consisting of, say, 1 second of
> actual test (CPU time being burned) and 1 second of sleep gets sped up
> by, say, 10%, it means that the CPU time really got slashed by 20%,
> which is actually *more* significant than without the sleep.

But in

u=1.33  s=0.46  cu=62.27  cs=10.54  scripts=273  tests=21120

real    2m16.388s
user    1m3.610s
sys     0m11.000s

(which is from time ./perl TEST)

I was under the impression that real is wallclock, and the other two
are user and system call CPU time, not elapsed time.

> I think the most dubious element of the test suite as a whole is the
> large number of separate and relatively small scripts, the operating
> system (and filesystem, and networking, and ...) dependent startup
> time may taint the results quite a lot.

Lots of lovely perl symbols, all being put into symbol tables lots of
times, all being hashed each time?

If so, that might explain it unfairly favouring faster hash algorithms

Nicholas Clark

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