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

Re: 2% speedup with Duff's device

Thread Previous
Alan Burlison
January 23, 2001 09:45
Re: 2% speedup with Duff's device
Message ID:
Nicholas Clark wrote:

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

User and sys time vary depending on what else is happening on the
machine - for example if the machine is busy, systime will tend to go up
because of extra context switches etc.

The testsuite is not suitable for performance measurement - too many
scripts, too much system interaction and way to many fork/execs.

I actually think there is a need for two types of performance tests,
overall tests capable of simulating 'real' perl scripts (v. difficult),
and scripts designed to test as small a subset as possible, e.g. hash
table access, scalar creation, maths ops, method calls, parse overhead
etc etc.  Each test should test (as far as possible) just one feature,
and should run for at least 1 elapsed minute, preferably 2-3 minutess. 
Glue enough of those together and you have a performance regression
testsuite ;-)

Alan Burlison

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