The Open University IMPACT Cluster "crunching the numbers"

We now have the latest version of theIntel Cluster Studio for linux
installed system wide. This will be a replacement for version 9 and will sit
along side version 11 (which is only available on the cluster)

To use please source

Bash
source /csoft/intel-12/bin/compilervars.sh intel64

Tcsh

source /csoft/intel-12/bin/compilervars.csh intel64

For more information see
http://software.intel.com/en-us/articles/intel-composer-xe
and select the support tab on the far right

Intel® Cluster Studio XE cluster tools include:
Intel® Composer XE compilers and libraries

The Portland Group

We now have a dedicated AMD compiler available system wide, We have chosen the Portland group
PGI Fortran/C/C++ Server combined Fortran and C/C++ Version 11.10-0

To use source
32 bit Bash
/csoft/pgi/linux86/11.10/pgi.sh
64 Bit Tcsh
/csoft/pgi/linux86/11.10/pgi.csh

To use PGI MPI source

32 bit Bash
/csoft/pgi/linux86/11.10/mpi.sh
64 Bit Tcsh
/csoft/pgi/linux86/11.10/mpi.csh

For more information see
Docs http://www.pgroup.com/resources/docs.htm
Compiler user guide http://www.pgroup.com/doc/pgiug.pdf

The 2011 christmas shutdown is now finished

IMPACT-STORAGE the main NFS server has been restarted after locking out all users.
the memory was doubled in an attempt to sort out the problems with increased ZFS file space and
more slots in the cluster using up all the available memory.

The JLB server room and air handling units are now protected by a backup generator
This was installed week ending 31st July 2011 after a long campaign by MCT sys admin Allan Thrower.
We thank him here for his sterling work in getting this through to completion.
All users of the IMPACT cluster will benefit from the extended run times this unit now allows us to provide.

IMPACT cluster new job script requirements & Quotas
Because recently nodes have been over subscribed and had the memory exhausted, 2 new forced parameters have been added to the default job script they are virtual_free & h_vmem please add them to the end of the line in your script that starts with
# job run time hours:min:sec (always over estimate )

#$ -l s_rt=25:00:00
to now read
#$ -l s_rt=25:00:00,virtual_free=4G,h_vmem=6G
mem_free=MEM_NEEDED,h_vmem=MEM_MAX
All jobs are now set to use 1Gb by default if you need less please request less
mem_free

IMPORTANT message about IDL use on the cluster at the end of this email

In the last week Allan and i have install 2 new machines into the cluster.
We received one from funds donated by the Science Dept and the other is a machine we were given free by Dell to test as part of our current expansion plans, this will not influenced the final choice in supplier, our choice will be based on "bang for buck" and support.

Allan has received the go ahead to start the procurement for the backup generator from Estates

Imp202 has now been repaired and is back in service.
The fault was traced to one of 2 memory modules both of which were replaced by our supplier.
As per a request by several users the old default queue for this node was removed and
and 2 new queues created one for large memory short term use (7 days only) 7 actual slots
and one for interactive use using qlogin 6 actual slots.

Usage of the queues will be monitored and amended where needed.

Please note:-
Starting Friday 11th March 2011
a script will be run automatically every friday on the /scratch area and will remove all files
and folders that have not been accessed in the last 30 days.
/scratch is meant as a dumping ground for running jobs and should not be used to store
data, jobs running more than 30 days will not be affected as long as they write to scratch at least once in the 30 days

May i please remind all users direct ssh access to internal nodes to run jobs is not allowed
any job found not running under Grid Engine control will be killed.
If users need interactive access to a node then qlogin must! be used and only one instance of a job should be run under that login