hg1 cluster

This page contains a short description of the HPC computational facilities available:


QUEUES SUMMARY

  • nehalem
        8 + 10 nodes, 2 x quad core Xeon E5520@2.27GHz, 12GB, infiniband
        suggested use    : MPI over IB ; large memory
        walltime limit   : 12 hours
        CPU number limit : 18 nodes (144 cores)
        suggested qsub   : qsub -q nehalem -l nodes=...:ppn=8 ...
  • zebra
        22 nodes, 2 x quad core Xeon, 16GB, infiniband
        suggested use    : MPI over IB ; large memory
        walltime limit   : 12 hours
        CPU number limit : 16 nodes (128 cores)
        suggested qsub   : qsub -q zebra -l nodes=...:ppn=8 ...
  • blade
        88 nodes (more or less), 2 x dual core Opteron, 8GB, infiniband
        suggested use    : MPI over IB
        walltime limit   : 12 hours
        CPU number limit : 64 nodes (256 cores)
        suggested qsub   : qsub -q blade -l nodes=...:ppn=4
  • iblade
        56 nodes, 2 x dual core Opteron, 8GB, infiniband
        suggested use    : MPI over IB
        walltime limit   : 12 hours
        CPU number limit : 64 nodes (256 cores)
        suggested qsub   : qsub -q iblade -l nodes=...:ppn=4
  • smp
        12 nodes, 2 x   dual core Opteron, 8GB
        suggested use    : serial or smp
        walltime limit   : 96 hours
        CPU number limit : one node (4 cores)
        suggested qsub   : qsub -q smp -l ncpus=...

FILESYSTEMS SUMMARY

/u

        home filesystem
        small, slow, hopefully quite safe; quotas are enforced
        suggested use: permanent and/or precious data

/scratch

        "traditional" scratch filesystem
        large, faster than /u on infiniband nodes
        suggested use: work and/or temporary data

/lustre

        parallel scratch filesystem
        large, much faster than /scratch on infiniband nodes
        suggested use: work and/or temporary data

/local_scratch

        local filesystem
        limited in size and not especially fast, but you are the only user
        available only on a*, i*, n* and p* nodes (smp, iblade, nehalem and zebra queues)

ACCOUNT EXPIRATION

All accounts expire after one year. At that time if you still have a valid SISSA or Democritos account, just write to helpdesk-hpc@sissa.it and ask for account extension.