CERN Accelerating science

Previous:		HPPLUS: a Change of Hardware	(See printing version)
Next:		The End of SP2 and SP1 Services

CSF, the Central Simulation Facility, started in 1991 as a service for compute intensive jobs with low I/O requirements. It was designed around clusters of HP 735 workstations with NQS (Network Queueing System) as the batch system. The addition of HP712 workstations in 1996 doubled the CPU capacity but the configuration has been unchanged since then.

Although the CSF service has been very reliable, there are a number of problems with the configuration today. These include

• insufficient disk capacity,
• lack of support for NQS,
• difficulties in maintaining the clusters, and
• relatively expensive management costs.

We are therefore taking advantage of the HPPLUS upgrade to move towards a new CSF service which is better integrated with today's batch and interactive services. The new CSF service will offer

• standalone nodes (recycled from HPPLUS),
• LSF instead of NQS,
• a larger working directory for the batch jobs (500MB instead of 200MB),
• AFS home directories and passwords, and
• job submission from all the public services (HPPLUS, RSPLUS, DXPLUS) (and, therefore, interactive logins will no longer be allowed).

One new CSF node is already available for test and more will be introduced in the coming weeks. The two systems will run in parallel until Easter 1999, at which point the current CSF service will stop. CSF users who would like to make early use of the LSF based service should send an E-mail to CSF.Support@cern.ch.

For matters related to this article please contact the author.