Bulletin 143 - 2005 Aug 11

1. HPCCC SX-6 SUPER-UX upgrade to R15.1
2. HPCCC SX-6 environment for SUPER-UX R15.1
3. WWW updates: SX-6 cluster local Userguide
4. cherax, CSIRO Data Store, and www.hpsc.csiro.au downtimes
5. TX7 software upgrade, and a new version of rsync on the TX7s

Node sx618 was upgraded to SUPER-UX R 15.1 on Wednesday 10th August.

Manuals for SUPER-UX R15.1 are being prepared for publication on the HPCCC WWW site.

Please note that the dates shown at
http://www.hpccc.gov.au/hpccc/userdocs/SuperUX_Node_Upgrade_to_15-1.shtml
are target dates for the changeover, and the actual changeovers will be staged: for example, for the move targetted for 17th August, the work will start on 16th August, and nodes will successively become unavailable until moved to the new operating system.

The move to R15.1 requires a significant amount of work on each node (including a disc re-configuration), and within the Internode Crossbar Switch, to isolate MPI traffic to one operating system level at a time.

Users please note - you do not have to make any changes to your scripts to continue using the R13.1 nodes, and should have no interruption to your SX-6 service. As the R15.1 nodes become available you will be informed and requested to test your applications by appending a "3" to your queue name.

In scanning the release notes, we have not found any changes (apart from bug fixes) which we believe will have an impact on users' work on the SX-6s.

Some of the R15.1 features are already in operation at this site, e.g. NQS II and ERS II enhancements.

The sxcross command now provides access to the SUPER-UX R15.1 libraries and environment, including MPI libraries.

Note that executables linked with the R15.1 crosskit and MPI libraries should not be run on a R13.1 system - all other combinations are supported.

To switch crosskit versions on gale/cherax/farrer you can do the following:

 sxcross_upgrade

this is an alias or function which runs: sxcross crosskit/r151 (you can also run that directly)

If you need to switch back to the default do:

 sxcross_downgrade

which runs: sxcross crosskit/inst

Run 'sxcross' with no arguments for an explanation and run sxenv (and env) if you need to check on your current environment.

Note that you need to do the following to make sxcross, sxcross_upgrade and sxcross_downgrade available:

• ksh: . /SX/local/etc/sxcross.sh
• csh: source /SX/local/etc/sxcross.csh

You may already be doing this - only the definition of sxcross_upgrade and sxcross_downgrade are new. They will only be available while there are multiple crosskits (for both SUPER-UX 13.1 and 15.1) and will be taken away when the transition to 15.1 is complete.

The SX-6 Cluster local Userguide at http://www.hpccc.gov.au/hpccc/userguides/sx/ and http://intra.hpsc.csiro.au/userguides/sx/ has been updated to include information on the upgraded sxcross command. See the previous item for more details.

The Userguide now includes a changelog section, and it is possible to view a marked-up version of the Userguide showing changes from the previous version.

Your comments on this feature would be welcomed, before we decide to make similar features available for other Userguides.

Phil Tannenbaum will present "Things You Never Wanted to Know about the SX-6, But Which Are Important Anyway" on Thursday 18th August, 11.00am, BMRC Seminar room, east side of 9th floor, 700 Collins Street

Abstract:

The SX-6 is a very different machine from the SX-5. Some of the issues BMRC users have been focusing on have been memory, I/O, and system scheduling. The presentation will present a major characteristic of SX-6 memory that is important to understand, and a discussion of how the SX-6 does I/O to both local and GFS file systems. A short discussion of system scheduling will be included.

Critical parts of the software for the scheduled upgrade on cherax did not get incorporated into SuSE SLES9 SP2.

The major software upgrade has thus been postponed.

However, the service interruption will still go ahead this coming Saturday 13th August, to allow the new disc to be incorporated into /cs/datastore and /work, and to upgrade the batch system, torque.