CSAR Bulletin issued in June 2002

Service Developments

As described in the latest edition of CSAR Focus, CSAR is currently installing new facilities to enhance many aspects of the service.

These include a new interactive machine, 'Wren' and new high performance disks via a storage area network. In addition, we are also in the process of replacing NQE (Network Queuing Environment - the batch job queuing system) with LSF (Load Sharing Facility - produced by Platform Computing) on the SGI Origin Systems.

Wren: A new Origin300, Wren, which has 16 500MHz processors, will replace Fermat as the interactive system. It will be purely for interactive (and administrative) work, thus providing more processing power for interactive use. Interactive access to Fermat will be withdrawn.

Storage Area Network (SAN): The SAN will provide a major enhancement to the disk provision and performance for Fermat, Green and Wren. The primary benefits will be improved access to home and /hold filestore from Green, thus removing problems experienced by some users with large I/O requirements. There will also be a globally accessible 'temp' area, files on which can be accessed from all Origin systems without requiring interactive access to the batch systems.

At present there is only one type of disk available for use on Fermat and Green - known as Medium Performance disk. With the SAN there will also be High Performance disk (as is currently available for Turing), Ultra High Performance disk, which will be faster still, and High Volume disk, which will provide a cheaper storage facility but will not be backed up. Existing data will remain on Medium Performance disk but you will subsequently be able to trade resources and use any of the 4 types of disk. In addition to the globally accessible 'temp', there will also be local temporary areas on both Fermat and Green. Installation is progressing well and it is hoped to have the new facilities available during June.

Use of /tmp on Fermat and Green: Please note that as part of the SAN installation process it will be necessary to rearrange the data held in the local /tmp directories on Fermat and Green. It is hoped that this can be achieved without any loss of data but this cannot be guaranteed. You are reminded that all data in such temporary areas is always at risk, never backed up and thus should never be relied upon for anything other than truly temporary files. Please ensure that you keep a copy of important data in home or hold areas. Please contact the CSAR Helpdesk if you have any queries on this subject.

Load Sharing Facility (LSF): LSF is currently being installed on Fermat, Green and the new Origin 300, Wren. This will provide better control for running parallel jobs than NQE, and in particular should minimise the extent to which resources such as memory are shared between jobs, thus improving job performance.

LSF does use different commands, options and job submission scripts from NQE, but the changes are generally straightforward and documentation will be provided to minimise any disruption during the transition period. However we urge anybody who may use NQE commands or scripts in an unusual way (such as embedding NQE script details within programs or other scripts), to contact us as soon as possible so that we can ensure that such cases are tested before LSF is made live.

ScaLAPACK

ScaLAPACK version 1.7 is available on Turing and versions 1.6 and 1.7 are now on Fermat.

All are located in /usr/local/unsupported/lib