[Bioperl-l] Bio-Perl workload for benchmarking Perl
robert.bradbury at gmail.com
Tue Apr 6 15:01:11 EDT 2010
>From my experience with even old desktop machines (2.8 GHz Pentium
Prescott) it is hard to get anything that takes an hour (even Firefox
or Chrome build from source in a few hours). Creating genome indexes
usually only takes a couple of minutes. Single gene blast homology
searches (which are generally done in C, not Perl) of entire genomes
only take a couple of minutes.
If you want something that takes an hour its going to have to be very
wide or very deep (low homology searches?; lots of gene searches
across many genomes?) The only things I can think of that require
that much CPU are whole genome assembly (which I don't recall BioPerl
being designed to do) or perhaps chromosomal synteny searches across
multiple genomes (which I also don't recall BioPerl handling).
The problem that one has (since one is benchmarking Perl) is the
identification of CPU intensive tasks which have not already been
dropped down to the level of using C code instead of Perl. (Also as
an aside, one might cite the trend to move even the C code into the
GPU as I believe has been done for projects like SETI at HOME,
Folding at HOME, etc. If its CPU intensive it is likely to be recoded to
take advantage of the best available hardware).
If one is going to do Perl benchmarking one needs to identify those
applications which are best implemented in Perl and are not
easily/have not been re-implemented in C or driven down to the
hardware level. So one really needs suggestions from the user
community as to what kinds of work fit that description that one could
construct benchmarks around.
On 4/6/10, Steffen Schwigon <ss5 at renormalist.net> wrote:
> Robert Buels <rmb32 at cornell.edu> writes:
>> Steffen: what kind of ballpark runtimes do you want for each
>> benchmark? 1 hour? 12 hours? 1 day?
> 1h would be enough, I think.
> It should be just big enough to not be completed within seconds when
> switching from my everyday desktop to a fast 16+ core machine.
> So if in doubt, I would also take a look at the 12h run…
> Kind regards,
> Steffen Schwigon <ss5 at renormalist.net>
> Dresden Perl Mongers <http://dresden-pm.org/>
> Bioperl-l mailing list
> Bioperl-l at lists.open-bio.org
More information about the Bioperl-l