Running Batch Jobs on Manchester Suns

How the Batch System Works

Alessandra introduced a batch system into the suns. It restricts the number of jobs to 4 (1 per CPU).

It is not the same as the bsub system at SLAC/RAL, but still very simple. You have to go into your workdirectory (or wherever). You make a file called myjob containing just:

. /bfactory/profiles/group_sys.conf.sh
srtpath 8.8.0c SunOS5
cd /export/home/roger/tau/workdir
./betarun ./mydata.tcl

Then enter:

   qsub -j oe -Nmyjob.log myjob

The -j switch merges cout and cerr output.
The -N switch means that the log file is called myjob.log.onnn, where nnn is the ID assigned on submission.

This can all easily be incorporated into a simple script to submit a whole lot of jobs at once.

The only other thing you need is the xpbs command, which throws up a window for monitoring queues, job progress, etc.

Comments

GDL: At present the batch system gives priority simply in the order of job submission
- A user who manages to get 100 jobs in the queue will get all his jobs through before anyone else is allowed to get in.
GDL: At the moment, four jobs can run simultaneously.
- In case we get in a situation where two or more people wish to use it at the same time, there is a switch
```
 -W depend=afterany:<jobid>
 
```
  which holds a job in the queue until job <jobid> is finished. So kind users can control how many of the four streams they use by submitting chains where each job depends on completion of the previous one. ...