Bulletin 146 - 2005 Sep 09

1. HPCCC SX-6 system upgrade to SUPER-UX R15.1
2. HPCCC SX-6 environment for SUPER-UX R15.1
3. Cleanups of temporary file systems
4. CSIRO - New Appointment

The last of the SX-6 nodes running SUPER-UX R13.1 will be upgraded from Tuesday morning 13th September.

Jobs and queues for R13.1 will be held back prior to the time of upgrade, but running jobs will be be lost if still running at the time of the upgrade.

After the final upgrade, all queues will be re-routed to run under R15.1.

Would users please remove references to the "3" queues after the final upgrade. The "3" queues will disappear on or after Wednesday 28th September.

There will be another SX-6 shutdown on the morning of Wednesday 14th September to allow the nodes to be re-bound into one cluster. There will also be a test of the backup power supply on that morning.

On Tuesday 13th September, the default cross-compilation environment will be changed to R15.1 on all cross-platforms.

If you wish to access the previous R13.1 environment after that, please use the command sxcross_downgrade and sxenv

to check the components. (Please do not use the old versions unless you need to for testing/troubleshooting. Note that sxcross_downgrade resets the compilers to default versions and downgrades the MPI to the R13.1 version.)

The commands sxcross_upgrade and qsub_upgrade will be withdrawn on or after Wednesday 28th September.

The HPCCC policy on file systems management for the SX-6/TX7 system is available at http://www.hpccc.gov.au/hpccc/userguides/sx/ under the heading "File Systems".

The removal of job and session temporary directories such as $TMPDIR, $LOCALDIR and $MMFSDIR on the SX-6/TX7 systems has not been happening automatically as thoroughly as advertised - many old directories have been left behind.

The HPCCC plans to implement more thorough measures soon, and clean up many old directories left behind - this will improve performance for some file metadata operations.

Flushing of $WORKDIR areas will be introduced in accordance with the policy set out in HPCbulls 133 and 132. Files from oldest to youngest will be flushed when space becomes critical, but no files newer than a particular threshold will be flushed - for some file systems, this limit is 30 days, for others it could be as low as 7 days.

On the CSIRO HPSC cluster burnet, flushing will be introduced on the $WORKDIR area in the next two weeks - when the file system reaches a critical level, files will be flushed from oldest to youngest, but stopping with files newer than a specific age. This will be set initially to 7 days, but may be varied as needed.

Important files should be copied from $WORKDIR to elsewhere.

CSIRO has announced that in his role as Director of CSIRO HPSC, Dr Rhys Francis will have additional responsibilities in CSIRO IT as the eScience Executive Manager from 4th October 2005.