Batch System Farm Altas Energias

Batch System Farm Altas Energias

Introduction

The Farm available for the High Energy Group IF-UASLP consists of several types of computers. hep00 is the main node, with disks and all software. 3 dual processor machines hep02, hep03, and hep04 are located in the Computer center and can only be accessed via the batch system. Other machines with monitors included are rutherford, cherenkov, charm, and sometimes fermi.

The batch system to distribute jobs to the different machines is called Fermilab Batch System New Generation (fbsng). Below we will show some examples how to use it. After a setup fbsng on whatever machine, you can see which nodes are up with fbs nodes, and the queues existing with fbs queues. To submit jobs, use fbs exec. A full user manual can be accessed here, and the there is a fbsng web site at Fermilab.

In general you need a script which will do the work for every job. The batch system will create two files in you home directory (*.err and *.out) containing the stdout and stderr of you job. It will also (optional) send mail to you.

In the following we will show two examples on how to use the farm in SLP.

SELEX Example - Running over the 1Million events

As first example we will run soap over the 33 files, which makes up the one million event sample of run 10783. I wrote a script to run soap, which does all the work. Note that the input filename is variable, so you can use it for all files. Obviously you have to adapt the script to your need, and you have to create to own soap.cmd and recdf.ocs files, and maybe your own soap.exe. To submit 33 jobs in one go, you say (with a little bash trick)
for i in `ls /usr/people/ehep1/e781/1M/*`; do fbs exec -q long -m jurgen `pwd`/soap_example.sh $i; done
and you wait. A full run will take about 5 hours.

CKM Example - Simulating the Pion RICH Resolution

For this example I wrote 2 scripts. One, as before, will do the work itself, and can be seen here. The other is to submit the jobs. We make a loop over momenta between 13 and 22 GeV, for both pion and muon in the pion rich. Each job takes up to 30min to run.


Last update at July 15, 2002 by Jurgen Engelfried