next up previous contents
Next: IMPORTANT NOTE ABOUT SCORE Up: Clusters of clusters Previous: sub clustering using multi-user   Contents

sub clustering using PBS

The file /var/scored/pbs/server_priv/nodes contains a list of the compute nodes available to PBS. The nodes are listed as hostnames together with s set of properties associated with each node. For example:
#
# `nodes' file for pbs_server, generated by scbd2pbs.
#
sc00.streamline          ETHER score  np=1
sc01.streamline          ETHER score  np=1
sc02.streamline          ETHER score  np=1
sc03.streamline          ETHER score  np=1
Here all the nodes have the property ETHER,score and np=1. The np=1 property indicates that these nodes have 1 virtual processors. The score property must be present in order to run SCore MPI jobs.

You may additionally define your own node properties by appending an arbitrary ascii string to the end of each line in the PBS nodes file. The PBS server must be restarted for changes in the file to take effect - e.g

[root@scserve nrcb]# /etc/rc.d/init.d/pbs_server restart
Shutting down pbs_server:                                  [  OK  ]
Starting pbs_server:                                       [  OK  ]
[root@scserve nrcb]#
For instance we could define the property bigmem on some of the nodes and maths (say) on other nodes giving in this example:
#
# `nodes' file for pbs_server, generated by scbd2pbs.
#
sc00.streamline          ETHER score  np=1 bigmem
sc01.streamline          ETHER score  np=1 bigmem
sc02.streamline          ETHER score  np=1 maths
sc03.streamline          ETHER score  np=1 maths
After restarting the pbs_server users will then be able to submit jobs like:
qsub -l nodes=4:score:bigmem
qsub -l nodes=1:maths
To submit a job on the bigmem nodes and maths nodes respectively ( here a 4 node SCore MPI job and a scalar job).

The PBS system is extremely configurable. For instance you can configure multiple queues e.g for short daytime jobs, long overnight jobs, jobs that require large memory etc. There is insufficient space to detail everything. Please see the PBS Administrator guide for full details. You can find the Postscript and plain text version in

/opt/score/src/pbs/src/pbs_admin_guide.ps
/opt/score/src/pbs/src/pbs_admin_guide.txt
respectively on the front end server.

Our Technical support team will also be happy to help you in customising PBS to your requirements.


next up previous contents
Next: IMPORTANT NOTE ABOUT SCORE Up: Clusters of clusters Previous: sub clustering using multi-user   Contents
2004-06-17