Hello everyone, First of all thanks to the developers for improving the installation of rosetta_refine. In contrast to my experiences half a year ago, this time the installation was easy and straightforward. However, refining a 600 Aa @ 3.5 A with 4 mol per asu one more than one processor still completely breaks my computer. (Intel i5 2500K / 7.8 GB RAM) I think this might be a memory issue. Having only a "normal" PC, one process already uses 5GB of RAM which equals 60% of my RAM. With only one process running the first of five model was finished in roughly 3 hours. However, the second model already takes more than 10 hours, and the log file on the _ros_tempXXX file wasn't update for the last 9 hours. I can't see any progress, however it still uses 100% of one processor. Unix's top tells me that "reduce" is completely using this resource and that it had run for the last 821 minutes... so I guess rosetta_refine is somehow stuck? Any ideas why it is stuck and how to solve that? With kind regards, Jan Gebauer By the way: Is rosetta_refine meant to work on a "normal" Computer - like mine. In principle I would have access to a cluster, but set-up time there would be considerably long for me... -- Dr. Jan Gebauer AG Prof. Baumann Institut für Biochemie / Uni-Köln Otto-Fischer-Str. 12-14 / 50674 Köln Fon: +49 (221) 470 3212 Fax: +49 (221) 470 5066 http://px.uni-koeln.de/
On Wed, May 7, 2014 at 11:56 PM, Jan Gebauer
With only one process running the first of five model was finished in roughly 3 hours. However, the second model already takes more than 10 hours, and the log file on the _ros_tempXXX file wasn't update for the last 9 hours. I can't see any progress, however it still uses 100% of one processor. Unix's top tells me that "reduce" is completely using this resource and that it had run for the last 821 minutes... so I guess rosetta_refine is somehow stuck?
This sounds like a bug in Reduce/phenix.refine - which I thought we had fixed already. Could you look in the run directory and see if there are stdout/stderr files corresponding to this job? I think I know of a workaround we can add for this but it will require modifying the Rosetta source code. By the way: Is rosetta_refine meant to work on a "normal" Computer -
like mine. In principle I would have access to a cluster, but set-up time there would be considerably long for me...
A cluster or large multiprocessor is strongly recommended - due to the stochastic nature of Rosetta's optimization process, you really need to sample multiple runs to get an optimal result in most cases. If you use one of the queuing systems we support (SGE, PBS, LSF, others that I can't remember) it should be able to automatically parallelize across cluster nodes, although I've only tried this with SGE. -Nat
participants (2)
-
Jan Gebauer -
Nathaniel Echols