Re: http://lists.bx.psu.edu/pipermail/galaxy-dev/2012-June/010153.html On Fri, Jun 15, 2012 at 4:52 PM, Peter Cock <p.j.a.cock@googlemail.com> wrote:
On Fri, Jun 15, 2012 at 4:38 PM, James Taylor <james@jamestaylor.org> wrote:
This is exactly what I think we should do (and have for a long time), but I think the variable should be something like:
GALAXY_CPUS
(threads is not accurate, a multithread or multiprocess job might want to use this info, something even more abstract than CPUS might make sense, but SLOTS has never made sense to me).
I agree that a Galaxy specific name makes a lot of sense, and that the SGE term "slots" is a bit odd. Using CPUS however is potentially ambiguous with CPUs vs cores - my desktop has two quad core CPUs, i.e. 2 CPUs but 8 cores.
Where do you think this number should come from? A new entry in the runner URL is simple albeit potentially redundant with cluster-specific entries in the runner URL. As to the alternative (doing it automatically), for PBS and SGE determining the number of cores from the cluster configuration and/or parsing the cluster runner URL sounds doable - what about the other backends?
Peter
Has the Galaxy team had any further thoughts on this topic? i.e. providing an environment variable or cheetah variable for the use of tool authors to set the number of threads/CPU cores to use. (With the value ideally coming from a default setting unless over-ridden via the [galaxy:tool_runners] entry in universe_wsgi.ini for that tool.) Thanks, Peter