Very Large Memory Nodes on the NIH Biowulf Cluster (Biowulf)

In order to meet the increasing demand for large memory
systems from genome assembly programs and other memory-
intensive applications, the Biowulf Staff is pleased to
announce the addition of very-large memory nodes to the
cluster.

Sixteen nodes with 72 gigabytes (GB) of memory each, as
well as a single node with 512 GB are now available.

Additional details on the hardware configurations
are available on the Biowulf web site:
  http://biowulf.nih.gov/user_guide.html#hardware

Allocation policies for Biowulf Very-Large Memory Nodes:

The 72 GB nodes will be dedicated to single jobs as is
the case for most nodes in the cluster.  Jobs submitted
to these nodes should require a _minimum_ of 8 GB of
memory.

To submit a job to the 72 GB nodes, specify the "g72"
property to the qsub command, e.g.:

qsub -l nodes=1:g72 mybatchjob

Note that while these nodes have 8 processors, your job
can use fewer than 8 and 'waste' the other remaining
processors, as long as more than 8 GB of memory is
utilized.

 The 512 GB node will be shared amongst multiple jobs.
Only jobs with memory requirements of greater than 72
GB should be run on this node.  This node also offers
a high speed local scratch space of 3 TB.

To submit jobs to this node, you must specify the job's
memory requirement in the qsub command, e.g.:

qsub -l nodes=1,mem=128gb mybatchjob

Also note that there is a 72 hour wallclock time limit on
jobs submitted to the 512 GB node.  Jobs exceeding this
time limit will be terminated by the batch system.

Finally, these allocation policies are subject to change
depending on the mix of jobs submitted to the large memory
nodes.

Please send questions and comments to staff@biowulf.nih.gov