Farm Usage

Advanced Research Computing is currently operating two farms, the parallel farm and the small-batch farm. These are both accessed from a single front end legion which is registered in both the its.unimelb.edu.au and the hpc.unimelb.edu.au domains (the farm nodes are only in hpc.unimelb.edu.au). HPC users who have satisfied the RAS criteria and been given an account on legion are able to use either farm. legion and the two farms share a common file system (via NFS) hence user files are visible on all nodes. Whilst the user discs are shadowed to protect against hardware failure, no other backups are kept. Users should use the disc space on legion only for working files and move important results, etc. back onto their home systems for backup and archive purposes.

The parallel farm is currently run on a lassez-faire basis. Parallel jobs can be run using MPI, PVM or HPF (GAMESS jobs are a special case), each of which uses a slightly different method. Users may also write shell scripts to implement trivial parallelism (where the same jobs is run many times with different data). Each node has 1.5 GByte of local scratch space (/scratch), and is run as a real memory machine, i.e. jobs can only start if there is sufficient memory available, paging is not allowed as it detracts from the maximum performance.

• MPI jobs
  We have installed and support the public-domain MPI implementation mpich. This is installed in /usr/local/mpi and prospective users should add source /usr/local/mpi/mpi.cshrc to their .cshrc files. More details on using MPI can be found in the MPI Usage page.
• PVM jobs
  We also run and support PVM (public-domain). This is installed in /usr/local/pvm3 and users should copy an appropriate .stub file from /usr/local/pvm3/lib and add it to their own startup file. More details on PVM can be found in the PVM Usage page.
• HPF jobs
  HPF is supported via the Compaq Parallel Software Environment (PSE). This requires the user to add
  

  		setenv PSE_FARM itsfarm
  		setenv PSE_PARTITION compute
  

  to their startup file. The use of PSE is signalled to the fortran compiler by the -wsf [n] flag. When a job compiled with this flag is run, PSE automatically starts tasks on the other nodes. More details can be found on the HPF Usage page.

The Small-batch farm is run under the PBS queueing system. Each node has a run queue with a job-limit of 3 and a user-limit of 1. These queues are fed from a single queue on legion. As this is the default queue users simply need to qsub batchfile on legion and their job will run on the remote machine. More details can be found in the NQS Usage page.