Recently, NAS systems engineers augmented 21 Westmere nodes with additional memory. Previously, all Pleiades Westmere nodes had 24 GB of memory per node, so a job that required more than 24 GB/node would not run. Now, 4 nodes are equipped with 96 GB/node and 17 nodes have 48 GB/node. The availability of these Westmere "bigmem" nodes will enable running many jobs that could not run before.

The method for accessing these nodes is similar to that for accessing the Harpertown bigmem nodes  (64 nodes with 16 GB/node). See "How to Get More Memory for Your Job."

You can access the Westmere bigmem nodes by using the :bigmem=true attribute in the PBS resource request:

 #PBS -l select=1:bigmem=true:model=wes

If your resource request follows the above example, your job will be assigned to either a 48 GB or a 96 GB bigmem node, depending on availability. To explicitly request a bigmem node with 96 GB of memory, add the :mem attribute with a memory size between 48 and 96 GB. For example:

 #PBS -l select=1:bigmem=true:mem=90GB:model=wes 

If you have an MPI job where rank 0 does the I/O and needs between 22.5-48 GB of memory by itself, you can request 1 Westmere bigmem node for rank 0. Consider the following example of a 48-process MPI job:

 #PBS -l select=1:ncpus=1:mpiprocs=1:bigmem=true:model=wes\ +3:ncpus=12:mpiprocs=12:model=wes+1:ncpus=11:mpiprocs=11:model=wes 

In the above, 1 bigmem Westmere node and 4 regular Westmere nodes will be assigned to this job. The bigmem node will be used by rank 0. Three regular nodes will run 12 MPI processes each, and one more regular node will run 11 MPI processes.

NOTE: These bigmem nodes can be used for jobs requesting any queues (normal, long, debug, etc.) except the devel queue.

If you have questions, please contact the NAS Control Room at (800) 331-8737, (650) 604-4444, or support@nas.nasa.gov.