Mahmood Naderan
2016-12-10 17:18:47 UTC
Hello,
Here is the problem with Torque/PBS job manager. I ran the following command to allocate 8 cores on a node, however, I myself know that node has less than 8 idle cores. As you can see, the interactive job started means that everything is normal!
***@cluster:~$ qsub -X -I -l nodes=compute-0-1:ppn=8
qsub: waiting for job 1488.cluster.edu to start
qsub: job 1488.cluster.edu ready
***@compute-0-3:~$
However, on another terminal, the output of 'qstat -an' shows that it actually allocated 4 cores
1488.cluster.edu mahmood default STDIN 5648 4 4 -- -- R --
compute-0-3/31+compute-0-3/30+compute-0-3/29+compute-0-3/28
Also, 'showq' command shows that only 4 cores were allocated
1488 mahmood Running 4 99:23:53:55 Sat Dec 10 20:34:10
Maybe this seems to be normal and the PBS intelligently, allocates the remanding cores. However, when the user submits an 8-core job, 8 processes are created on the node. That means, the PBS allocated 4 cores but the job is running on 8 cores.
Since the node hadn't 8 idle cores, such situation shows that the CPU utilization is less than 100% because some cores are running multiple processes.
Any thought on that?
Regards,
Mahmood
Here is the problem with Torque/PBS job manager. I ran the following command to allocate 8 cores on a node, however, I myself know that node has less than 8 idle cores. As you can see, the interactive job started means that everything is normal!
***@cluster:~$ qsub -X -I -l nodes=compute-0-1:ppn=8
qsub: waiting for job 1488.cluster.edu to start
qsub: job 1488.cluster.edu ready
***@compute-0-3:~$
However, on another terminal, the output of 'qstat -an' shows that it actually allocated 4 cores
1488.cluster.edu mahmood default STDIN 5648 4 4 -- -- R --
compute-0-3/31+compute-0-3/30+compute-0-3/29+compute-0-3/28
Also, 'showq' command shows that only 4 cores were allocated
1488 mahmood Running 4 99:23:53:55 Sat Dec 10 20:34:10
Maybe this seems to be normal and the PBS intelligently, allocates the remanding cores. However, when the user submits an 8-core job, 8 processes are created on the node. That means, the PBS allocated 4 cores but the job is running on 8 cores.
Since the node hadn't 8 idle cores, such situation shows that the CPU utilization is less than 100% because some cores are running multiple processes.
Any thought on that?
Regards,
Mahmood