I have a machine (machine1) with 96 CPUs. When I submit an abaqus job directly on this machine, the calculation time is about one and a half hours. However, when I submit it to this machine (machine1) through pbs and use the same number of CPUs to calculate the same job, it takes two and a half hours. How can I solve this performance problem?
Please check the below while running without openpbs and with openpbs
- system environment variables
- application environment variable
If you could share the application batch command line used with and without the scheduler.
But，I’m sorry, I don’t know which environment variables I should check,could you give me some hints?
Please share the process of submitting jobs directly on that 96 core host, before doing that type “env” and get the enviorment variable setup and the batch command line used
Please share the script and the process of submitting job through PBS. If you are using a script , add a line “env” to that script , just before calling the batch commandline
I used a machine with 18 CPUs to calculate a job in same situation. The job was completed about half an hour faster when submitted directly or through pbs. The details are as follows:
Thank you @wakaka for sharing the details.
create the below script in the /db/hq-throughpbs folder and submit the job
Plesae make relevant changes with the input file.
#PBS -N hq
#PBS -q workq
#PBS -V
#PBS -l select=1:ncpus=18
#PBS -o /db/hq-throughpbs/std.out
#PBS -e /db/hq-throughpbs/std.err
cd $PBS_O_WORKDIR
env
ls -ltr
/opt/software/abaqus/Commands/abq6144 job=hq input=hq.inp cpus=18 int
I submited the job after changed the script, but the completion time didn’t improve. The details are as follows:
Could you please try interactive console job and run the batch command line and check whether it runs quicker
qsub -V -l select=1:ncpus=18 -I -X <press enter>
computenode$: cd /db/hq-throughpbs
computenode$:/opt/software/abaqus/Commands/abq6144 job=hq cpus=18 int