My cluster is running Sun's Grid Engine version "GE 6.2u5 $Date: 2009/12/01 12:24:06 $". I'd like to submit a single job to the queue which is defined by a bash script containing a number of commands. Most of the commands in this script can only make use of 1 CPU. However, one single command can make use of N CPUs. Is it possible to somehow request more processors for just this one single command in my script?
Dynamically allocating resources like this is impossible with SGE. The proper way to handle this situation is to split the script defining the job into three pieces: A,B,C where A=script with commands needing onlu 1 cpu B=script with command needing N CPUs C=script with commans needing only 1 cpu
Then the trick is to submit these three scripts as three different jobs with the final two being dependent on the previous two being completed.