develooper Front page | perl.perl5.porters | Postings from October 2016

Re: regarding slow tests

Thread Previous
Jarkko Hietaniemi
October 18, 2016 22:33
Re: regarding slow tests
Message ID:
> later I will follow-up with some more.

It must be later.  Some logs attached.

Contained are:
- three repeated logs from the recentish beefy imac (osx)
- three repeated logs from the recentish box booking offers (linux)
- one log from mid-range mid-age sparc (solaris)
- one log from mid-range mid-age mips (irix)
- one log from puny old-age vax (netbsd)

You can use the Porting/, or since these
have all the raw numbers, whatever you feel like (e.g. to find
out medians and other percentiles, which the does not).
Other ideas: compare across architectures.  Look for things
having low CPU but high wallclock: possible explicit sleeping
and/or blocking on waiting for something.

To give you some idea of what slow means, in this vax box: the harness
part of "make test" takes almost 26 hours to run.  Before that the
part of the make that prepares the extensions takes a long time,
I haven't measured it exactly (should insert some "date" calls),
but it could be somewhere in the neighborhood of half an hour.
Maybe an hour.  Who's counting? The slowest tests take up to 1.5
hours to run.

If you don't have a vax lying around, but you want to feel the
blistering lack of speed, a modern equivalent could probably be
an old raspberry pi.  Don't get a fast one.

The names of the log files show the CPU speed: the vax has 72MHz.

I have no proof but I have a theory that part of the extreme
slowness could be not just the x35 CPU speed difference, but
also the much slower I/O.  Many of the tests also do a lot of
starting of processes (XS-APItest, ExtUtils-*) which also seems
to be horribly slow.

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