Thanks for your replies Nat and Jon,
Given the answer of Nat, phenix.phaser is not (by default) OpenMP, which
explains the non-parallel behaviour of my phenix.phaser.
I think I have been misleading by the GUI behaviour that actually show 3
threads for python (1 for phenix and apparently -for some reason- 2 for the
phaser gui, one of those disappearing quickly).
As I want to run phaser with a few different model, but as quick as
possible, I switched to the ccp4 phaser that parralelize and run well on
the cluster.
Best regards,
Lionel
2013/7/10 Nathaniel Echols
On Wed, Jul 10, 2013 at 7:16 AM, L. Costenaro (IBB)
wrote: I am trying to run phaser in a SGE cluster (qsub) using multiple proc (either phenix.phaser from the command line or phaser-MR from the GUI), but the jobs do not parallelize. When I run the phaser-MR locally (same executable) it does parallelize (multiple python threads).
I'm a little confused - what parameters are you using and what behavior do you expect? As distributed, Phaser itself (i.e. the Phaser-MR or Phaser-EP GUIs, or phenix.phaser) will not run in parallel, regardless of whether you're running the jobs on a cluster or locally. If you want to use the OpenMP compiler extensions to parallelize single jobs, you need to build from source *without the Phenix GUI*, because OpenMP does not play nicely with Python multiprocessing (or vice-versa, I guess). I'm not sure what makes you think it's being run in parallel locally, but I suspect you're being confused by the fact that executing a job from the GUI starts a separate process.
The MR programs that run in parallel are a) phaser.MRage, Gabor's automated system which is designed to try many different models (including different model preparation protocols), and b) my somewhat obsolete "parallel Phaser" GUI, which is a much more minimalist approach. These can both execute on a cluster, but the individual jobs will still be single-processor. (Currently both are still flagged as "alpha" features in the GUI, although MRage will hopefully be a fully supported program in the near future.)
-Nat _______________________________________________ phenixbb mailing list [email protected] http://phenix-online.org/mailman/listinfo/phenixbb