Connecting Infrastructure, Connecting Research

Glasgow Computing Service

Glasgow Computing Service provides access to 34 nodes. Each node has dual quad core Opteron 2384 processors and upto 32 GB of RAM. The NGS has been allocated 10% on this resource using Fair Share. Jobs are scheduled on the service using Torque/Maui.

Service Level Description

All users registered with the NGS have access to this resource using the accepted Globus-based access mechanisms (GSISSH, GRAM, GridFTP).

To access the resource using GRAM, GSISSH or GridFTP, use the hostname headnode01.cent.gla.ac.uk. The GSISSH service listens on port 2222. If your local installation of gsissh does not default to this port, you can connect with the command gsissh -p 2222 headnode02.cent.gla.ac.uk

To access the front-end node via GRAM, use the resource manager name headnode01.cent.gla.ac.uk/jobmanager-pbs.

The jobmanager-fork maps onto the jobmanager-pbs.

The home directory for each user is shared and consistently named across the head node and all execution nodes in the Glasgow Computing Service system. Accounts are allocated via the standard globus pool account mechanism, and active accounts which have not been accessed in the last 7 days by one of the standard grid access mechanisms listed above will be cleared and added back into the accounts pool. Active accounts will have their .globus subdirectories automatically purged of any file greater than 7 days old.  

In addition to the home area a temp area is available to the running job on the execution node – it can be accessed using the path   /tmp/$PBS_JOBID. This area is cleaned up at the end of the job – so it is the responsibility of the users script to copy any relevant data back to the home area before the job finishes.

Compiling

Standard GNU/Linux compilers are present on the system (e.g.  /usr/bin/gcc).

If you need to compile your code against the installed packages – gsissh into the headnode and call up an interactive batch job with qsub –I (the headnode has additional packages installed and if you compile on the headnode you may find that the job will not run on an execution node).

 For MPI – mpich is installed at /opt/mpich/. The installation of torque is mpich aware and mpich will get its node list from torque. See http://www.gla.ac.uk/services/it/forresearch/computecluster/userguide/howtorunmpijobs/

 Job Submission

Jobs can be submitted through either the standard globus job submission tools, or submitted manually. Instructions for manual job submission can be found on our main HPC website.

Contact Information

User support requests for Glasgow Computing Service should in the first instance be directed to the NGS Support Centre: support@grid-support.ac.uk