Parallel computing in RosettaDock3.1?

[Anonymous]

I’m trying to compile rosetta3.1 from source with options:
[…]#python scons.py bin mode=release extras=mpi

My task is to use RosettaDock3.1 to generate 4000 decoys of each one case in docking benchmark 2.0 and the number of the processors in amd cluster is more than 64. The scheme is that each time when I qsub a job, it can be partitioned into ten parts with each part returned with 400 decoys. The option -out::prefix is added in each part to distinguish the decoy for the reason which processor produces.

Before implemention, I manually create ten folders as the working directory with each counterpart with one of the ten parts. How can I code the script in PBS? Can I have a copy of the script via mail at lxhlhy@yeah.net or xhluo@ivt.edu.cn.

Thanks & Regards, xhluo

[Anonymous]

What are you asking for…? It sounds like you know what to do so I’m not sure what you need help with.

[Anonymous]

Let target 1acb in docking benchmark2.0 as an example. I want to get 4000 decoys and there are more than 64 processors in amd cluster. How to let 10 processors work and every processor returns 400 decoys?

Thanks & Regards, xhluo

[Anonymous]

A) Make the ten directories that the MPI docking executeable expects – they’re somehing along the lines of output_0 through output_9, or out_1 through out_10, I forget.

use the option -nstruct 4000

C) compile in MPI mode

D) Submit a job to your cluster that requests 10 processors and runs via mpirun, or whatever your equivalent is.

I believe the docking executeable’s MPI mode automatically assorts an equal number of jobs to each processor. I could be wrong; it could be a “pooling” implementation and it needs 11 processors to do this, where the 11th is a head node allocating jobs to the other processors.

Docking has been ported to JD2 for the next release, which makes all this much simpler (no need for multiple directories, control of jobs-per-processor, etc).

[Author]

Posts