I am trying to do parameter tuning of my learning model on a Bright compute Cluster, which requires a large number of jobs due to the number of parameters being tuned. Each combination of the parameters requires around 162 qsub jobs. And there are around 50 combinations of parameters that I require to check. This is equivalent to running around 162*50 ~= 8100 jobs. However there is a 350 qsub job limit per account on the cluster that I am using. I was hence wondering whether there was a way in bash scripting to check the number of currently active qsub jobs so I could effectively automate the process of initiating new jobs.
Asked
Active
Viewed 1,171 times
0
-
why don't you try qstat and qdel commands? – Sharif Mamun May 05 '14 at 06:56
-
1Have you tried qstat | grep
| wc -l? – dbeer May 05 '14 at 18:07
1 Answers
0
Did you already try with job arrays? You didn't specify the scheduler you are using (PBS, OGE, ...), but there should be a way to define a job array and, in the whole array, a limit on the number of tasks really running at a time. In PBS
#PBS -t 1-1000%100
creates a one thousand job array limiting to one hundred the number of tasks effectively running at a time. If you really want to find a way to check active jobs to automate the process of initiating new jobs, the qstat output should help you, but this should be the duty of your scheduler, not your.
deggio
- 31
- 3