National Computational Infrastructure
NCI National Facility
Newsletter December 2007

Table of Contents

Time allocations for 2008 The Merit Allocation committee met on Dec 5 and users will be informed shortly of their allocations for 2008. First quarter grants will be installed on January 1.

There were 142 application to the Merit Scheme including 29 new projects. The total time available in a year for the Merit Allocation Scheme is about 12.8 million SUs. The total of all requests for 2008 was about 19.5 million SUs. As a result it was not possible for all projects to be granted their full requests.

The Merit Allocation Committee judged the applications on the basis of how well they suited the MAS criteria and whether the current usage was efficient and productive. In some cases it was difficult to make this assessment as applicants provided too little or irrelevant information. It is in your interest to provide sufficient justification for your request so that a fair assessment can be made.

Holiday arrangements As the ANU will be closed from Christmas Eve till New Year's Day emails to help may not be answered immediately between Dec 22 and Jan 1. You may receive a reply if a staff member is on-line but do not depend on this happening. The official position on out-of-hours support is available here.
AC SLES10 upgrade The long awaited upgrade of the AC operating system to SLES10 Linux is finally underway. As compute nodes drain over the next couple of weeks we will reinstall them without noticeably disrupting the AC service. We apologise to those users who are and will suffer extended queue times of long jobs while this transition occurs.

Some of the changes that SLES10 brings are:

  • Numerous updated system packages. In particular, the GNU compilers and libraries are now based on version 4.
  • The ability to migrate the memory of jobs within a host. This should facilitate better scheduling and lead to reduced suspension times.
  • Fixes for several problems that have plagued a number of users.
  • Changing passwords on AC is now supported!
  • An updated MPI library with better support for oversubscribing cpus and threads as well cleaner MPI job aborts.
  • Generally improved memory management and cpu scheduling.
  • Some improvements in stability through better system management software.
  • An updated version of the cluster filesystem (CXFS) which should fix the issues that have required frequent reboots of the login node.
Note that because the underlying GNU toolchain is updated in SLES10, users will have to move to using version 9.1 or later of the Intel compilers for building applications. Version 9.1.051 has been the default Intel compiler version for some time so most users will not notice this restriction. Existing executables will continue to work.
AC Hardware reconfiguration During October, the AC was repartitioned into 64 cpu partitions (previously it had predominantly 32 cpu partitions). The main reason for this was to consolidate IO hardware and attach it to a smaller fraction of the AC's compute bricks. We have found that IO load can noticeably affect jobs on the cpus to which IO is connected. The reconfiguration has halved the number of cpus affected. A further consequence is that we can now run a few more parallel jobs at the same time. We have also consolidated jobfs disks allowing for the possibility of larger jobfs requests.
Increased access to bonus time This is a reminder of the new model for accessing bonus time on the National Facility systems. Previously, once a grant had been exhausted, jobs could only be submitted to the special restrictive bonus queue.

It is now possible to continue to submit jobs to the normal queue even after a project's grant has been exhausted. The priority of such jobs is reduced and they are more often suspended than other jobs in the normal queue. However, in general, the resource limits for the jobs will not be decreased. The impact of this bonus time usage will be monitored to ensure that projects that have not used their grant will still get priority access to the system.

Special cases for top-ups grants within a time period can still be made through the National Facility helpdesk.

Software update More software is continually being added to the AC. In particular, there are now more licenses for Mathematica and the Totalview debugger. The Intel compilers are regularly updated as new versions are released. The default version of the Intel compilers was recently changed from 8.1 to 9.1. Gaussian has been updated to g03 revision E.01.
Details on using different software is given at http://nf.apac.edu.au/facilities/software/.
Some packages have restricted access and these are marked accordingly and information on applying for access is given on the software web page.
Users are informed of new installations or updates in the message of the day at login.
Requests for new packages or updates can be made using the software request form.
NF Courses Staff of the APAC National Facility provide a range of training courses on using the National Facility machines and programming techniques. The full list is given on the training web page. These courses have been updated to be relevant to the Altix Cluster and several of them have been held recently at ANU, iVEC and SAPAC.

If you would like any of these courses given at a different partner site please contact us and it can be arranged. As these are hands-on courses we ask that a representative from the site arranges for access to a computer teaching lab with web access and a method of obtaining secure shell logins to the AC.

Email problems, suggestions, questions to