Notice of the start of vayu service

The NCI NF has now accepted the first stage of the new Sun service and is in a position to transition users of the existing AC over to the new service. Since the AC compute service will end around Sept 22, it is in your interests to make this transition promptly. Details of the interim system can be found at http://nf.nci.org.au/facilities/vu/hardware.php and usage is described in the NF User Guide at http://nf.nci.org.au/facilities/userguide/

To connect to the new system, known as vayu (or vu for short), you can: ssh username@vayu.nci.org.au with the same username and password as you have on other NF systems. You should find an environment very similar to other NF systems, particularly XE - the same basic filesystem structure, queue structure, etc.

Filesystems and Data

For your convenience, we have transferred your AC home directory as of midnight Thursday over to vayu. Some points to note:

  • if you have been active on AC this week, you should confirm that recently updated files have been transferred.
  • we have removed all object files and binary executables since they will not run on vayu; you need to recompile.
  • we have also renamed your AC shell "dot files" (.login etc) to a file with a "ac" prefix and installed system default versions. Since the names and versions of modules on vayu may be different to those on AC, we recommend you take some care in modifying these files.
  • home directories for those with early access have not been updated.

Please delete what is no longer needed.

Your /short directories have not been transferred. The AC login node and filesystems will remain available until at least Oct 5. If you need assistance to transfer relevant parts, contact help.

Resources

Users need to be aware that until the full Sun installation is completed later in the year, we only have 3GB per core of memory. XE has a small number of nodes with 4GB per core and as of next year vayu will have nodes with 6GB per core.

The configuration of jobfs on the new system is a challenge due to the new (modern) diskless compute node model. We are working through various options to find which is the most effective setup. This should be transparent to users. You should continue to use jobfs as before but, as always, you should request only what is really needed by your job. There may be issues of jobfs usage affecting your /short quota - bear with us while we resolve these.

Some users, hopefully most, will find vayu to be noticeably faster than AC or XE. We would welcome any feedback on relative performance of the various systems for your application. You should reconsider your walltime and ncpus requests in light of performance changes.

We have tried to migrate queue limit and disk quota extensions from other NF systems. Let us know if you are constrained by the limits you find.

In a followup email, we will clarify the charging regime that will be implemented for 2009Q4 and 2010.

Level of service

We expect there will be issues with the new system and in recognition of this, we will:

  • not be charging for usage of the new system before October 1.
  • Initially, hold jobs requesting long walltimes and release them when we are confident of the stability of the system.

Users with very long running jobs might want to consider using XE for the next week or two.

Please report any issues to help@nf.nci.org.au.