Contents
PROOF
The proofcluster.py script can be used to configure, start and stop a PROOF cluster on the batch system. The configuration is stored between sessions, so there is no need to configure again if the same configuration should be used again.
The following lines are an example workflow:
~samson1/proofcluster/proofcluster.pl config ~samson1/proofcluster/proofcluster.pl start ~samson1/proofcluster/proofcluster.pl status # do something with PROOF ~samson1/proofcluster/proofcluster.pl stop
SFrame
SFrame can be run locally, as PROOF Lite and full PROOF.
PROOF Lite
For testing purpose or short runs, you can run on the work group servers. For longer runs please use the batch system via qrsh using the multicore parallel environment. The following two examples should give you an idea how to do this.
Example 1: Request a session on the batch system, setup and run SFrame:
# get session on batch qrsh -pe multicore 8 -now n -R y -l site=hh -l h_cpu=01:00:0 -l h_vmem=1G # setup SFrame cd atlas/SFrame ini pyroot526 source setup.sh cd user/run sframe_main ../config/FirstCycle.xml
You can run another sframe_main in the same session.
Example 2: Setup SFrame and run it directly on the batch system
cd atlas/SFrame ini pyroot526 source setup.sh cd user/run qrsh -pe multicore 8 -now n -R y -l site=hh -l h_cpu=01:00:0 -l h_vmem=1G -V -cwd sframe_main ../config/FirstCycle.xml
When SFrame ends, you are back on the work group server.
Some comments on the qsub options:
- The option -R y tells SGE to reserve slots, even if not all will be directly available.
- As always for batch jobs you need to specify the needed CPU time and memory. Memory limits are given per core.
- You should specify a site using -l site=hh or -l site=zn, if you know where your data is. For Lustre the directory name contains the site, e. g. /scratch/hh. For dCache the DDM site name contains the site, e.g. DESY-ZN_LOCALGROUPDISK. If you do not care or do not know about the site, leave this option out.
- The option -V will import the full shell environment to the batch session.
- If option -cwd given the batch session will change to the current working directory. Please note that this needs to be either in AFS or /scratch.
PROOF
You can also run SFrame in full PROOF mode and get many more cores than in the PROOF Lite mode. Set up a PROOF cluster as described above. Add the PROOF master (hostname:port) to the cycle XML file under ProofServer. Then run SFrame as in local mode. e. g.:
cd atlas/SFrame ini pyroot526 source setup.sh ~samson1/proofcluster/proofcluster.pl start ~samson1/proofcluster/proofcluster.pl status cd user/run sframe_main ../config/FirstCycle.xml ~samson1/proofcluster.pl stop
Do not forget to stop your proof cluster, when done.