Commands for the PBS Queueing-System
The PBS queueing system integrates all computers in
the CIP-F room into one compute cluster. PBS schedules
and executes jobs across all of the cluster nodes. Jobs can be submitted into the queueing
systems from any node of the cluster. Important PBS
commands are:
qstat
gives an overview over jobs running on the cluster in a simple output
format:
Job id Name User Time Use S Queue
---------------- ---------------- ---------------- -------- - -----
1201.cicum81 Diani-Kat_Isobu c1165 126:46:0 R long
1227.cicum81 Didma-Kat_SiMe3 c1165 80:42:06 R long
1242.cicum81 Benzh_AllylTMS_ c1165 61:18:29 R long
1250.cicum81 py50ax zipse 01:34:42 R long
1251.cicum81 py51a zipse 01:25:50 R long
1252.cicum81 py51b zipse 01:19:16 R long
The first column designated with Job id displays
a unique identification number for each of the jobs such as "1201.cicum81". This job
carries the identification number "1201" and was submitted from node "cicum81".
An additional name provided by the user on submit time is displayed in the second
column designated Name. Only the first 15 characters of
the name are shown. The third column lists the job's owner. The fourth column
Time Use lists the elapsed CPU time for each of the
jobs in the format hh:mm:ss. The fifth column S describes
the status of a particular job using "R" for running jobs and "Q" for jobs that are queued
and waiting. The last column Queue describes the type of
queue the jobs are running in. Currently, there is a "long" queue available for
jobs exceeding 30 min CPU time and a "short" queue for shorter jobs. In case no
queue is selected explicitly, the "long" queue is chosen by default.
qstat -n
provides the same basic information as before and adds the requested memory and requested
CPU time (if set by the user on submit time) as well as the name of the node allocated to
a particular job.
Req'd Req'd Elap
Job ID Username Queue Jobname SessID NDS TSK Memory Time S Time
--------------- -------- -------- ---------- ------ --- --- ------ ----- - -----
1201.cicum81.cu c1165 long Diani-Kat_ 16420 -- -- -- 336:0 R 127:0
cicum86/0
In this particular example the job 1201 is running on cluster node "cicum86".
qstat -f
provides the full information for each job. The wealth of information provided by this
option is rarely necessary (or even helpful).
qsub file-name
submits a command file named file-name for execution
into the PBS queueing system. A command file can contain both
commmands for the queueing systems as well as an input file for a computational chemistry
package. An example for running a HF/6-31G(d) single point calculation with Gaussian 03 (g03)
can be found here
qsub -l nodes=node-name file-name
submits a command file named file-name for execution on a
particular cluster node named node-name. This can be
particularly helpful for queueing a sequence of jobs that have to execute
one after the other.
qsub -q long file-name
submits the command file file-name into the "long" queue. This only
makes sense if the queue has not been specified explicitly already in the
command file itself.
qdel job-id
deletes the job with job id job-id from
execution/the waiting list.
pbsnodes -a
gives an overview over the status of all nodes of the compute cluster.
Additional information is available in the PBS manual pages.
last changes: 16.10.2003, HZ
questions & comments to: zipse@cup.uni-muenchen.de