Bulletin 168 - 2007 May 10

1. Safe-keeping of Data
2. Super-UX Release 17.1 Upgrade: Aug-Sep 2007
3. Identify your slow SX-6 executables: sxcoffinfo -a
4. Delivery of NQSII & ERS "auto-generated" emails
5. New disc for cherax
6. cherax instability
7. SGI seminar - Michael Brown
8. APAC Grid Geosciences Roadshow

Note: "CSIRO" items can apply to BoM users of cherax and burnet

A user recently lost a large amount of work which was embedded in scripts, and stored in the $WORKDIR region of cherax.

This region is subject to flushing, but the user was un-prepared for this, presumably because the flushing has been fairly infrequent recently. Though we don't flush files younger than 7 days, still we advise against having any files in /work that can't be recreated because nobody knows when they might be called away to attend to more urgent matters.

The file /work/flush.status shows the flushing status, for which the last three entries recently were:

  Last flush of /work occurred Wed Apr  4 11:08:43 2007 UTC
  16270 files older than Tue Jan  9 06:03:27 2007 UTC were deleted
  10040 empty directories were deleted
  Last flush of /work occurred Wed Apr 25 15:48:16 2007 UTC
  50890 files older than Wed Feb 21 03:56:21 2007 UTC were deleted
  3966 empty directories were deleted
  Last flush of /work occurred Wed May  2 01:06:33 2007 UTC
  96507 files older than Fri Mar 30 10:07:48 2007 UTC were deleted
  198 empty directories were deleted

We need users to be aware of the management policies in place for each type of file system - see for example the User Guide at http://intra.hpsc.csiro.au/userguides/ds/ under the heading File Management.

Really important files that are hard to re-create should be kept at multiple locations. Note that the CSIRO Data Store does not keep data at more than one location.

[ page top ]

Major OS Upgrades are necessary every two years, to ensure that NEC can provide the best available support, in the event of critical problems. Improved and better integrated functions, features and bug fixes are gained, by keeping the OS relatively up to date.

Resolution of a large number of Bureau and CSIRO issues have occurred in the last 2 years. Similarly, compiler performance and executables are faster with the latest compilers and MPI/SX libraries. Hence, users are encouraged to check the compiler revision of executables and object files. MPI users will be required to relink their code with the latest MPI libraries, and execute with the corresponding later command and daemon.

[ page top ]

Executables from a compiler revision more than 2 years old, may require a recompile, rebuild and/or relink, to avoid problems with the latest OS and libraries. Hence, users are STRONGLY encouraged to check the compiler revision of executables and object files.

To obtain a report for your code, run -

"sxcoffinfo -a ./compiled_binary" or
"sxcoffinfo -a ./object_file.o"

For help on options, use -

"/SX/local/bin/sxcoffinfo -help" for gale/eccles/mawson/TX7
"/SX/local/bin/sxcoffinfo -help" for cherax

NOTE: Detection of any executables, compiled with the obsolete flag "-h sx4", will mean the code must be recompiled.

[ page top ]

Please see HPCbull 167 item 3 for background.

In the short term, NQSII and ERS automatically generated email will be centrally tagged for delivery to the recipient. In the longer term, NEC may enable inclusion of a site-defined tag, in the subject of the auto-generated email for these software products.

System Change Notice "2007-A11", on 20 April 2007, explains the site changes. See user news and advice, on the HPCCC website, at - http://www.hpccc.gov.au/hpccc/user_news_advice/

CSIRO HPSC have on order (lease) 22.5 Tbyte of cheap disc, and 7.5 Tbyte of high-speed disc to replace the main $HOME migrating disc space.

[ page top ]

For many of the latest crashes, the fault has been identified in the tape subsystem code, and after the crash on the evening of 3rd May, a correction has been applied.

Also, there was a small fault in DMF over recent weeks, which resulted in the second copy of about 400 files being wrongly written.

Work is underway to re-write the second copy of these files.

[ page top ]

"Accelerating Multi-Application Workflows or What Do You Do when CPUs are Free", a seminar presented by Michael G. Brown, Sciences Segment Manager, SGI.

Wednesday 23rd May, 14:30, at BMRC Seminar Room, 9E, 700 Collins Street, Docklands Vic.

[ page top ]

The APAC Grid Geosciences Roadshow will be a 2-day workshop to educate scientists and users in the geosciences community on grid- enabling their science. You will learn how the grid can aid you in your research and provide you with useful tools on accessing the myriad of geoscience software, computational and data resources on offer. The tools demonstrated will be the basis for NCRIS 5.13 Auscope research infrastructure.

The Roadshow will be in:

Melbourne 31th May - 1st June 2007 at VPAC, Melbourne

Brisbane 5th - 6rd June 2007 at ESSCC, University of QLD

Canberra: 27th - 28th June 2007 at Geoscience Australia, ACT

for more information see https://www.seegrid.csiro.au/twiki/bin/view/Compsrvices/APACGeosciencesProjectRoadshow2007

[ page top ]