Greetings Pleiades Users,

When the Pleiades system is returned to production, 32 new Westmere racks plus 20 Nehalem racks will be open to all users for running jobs. To take advantage of the Nehalem or Westmere nodes, you will need to make a change in the way you submit jobs. You will be able to specify which processor type you want your job to run on. There are three options:

• Harpertown (8 cores/node, 8 GB memory/node, SBU factor=0.9/core)
• Nehalem-EP (8 cores/node, 24 GB memory/node, SBU factor=2.2/core)
• Westmere-EP (12 cores/node, 24 GB memory/node, SBU factor=1.8/core)

Note that up to this point, unless you have been using the normal_N or normal_W queues, your Pleiades jobs have been running on Harpertown nodes.

The usage restrictions that were formerly in place for Nehalem and Westmere nodes will be lifted; users will be able to select any node type they wish. You will select which architecture to use based on a model attribute:

• model=har (Harpertown)
• model=neh (Nehalem-EP)
• model=wes (Westmere-EP)

For example:

#PBS -lselect=4:ncpus=12:model=wes

will route the job to the Westmere nodes. If you don't specify a model type, the default will be Harpertown.

If you have been running on Nehalem and/or Westmere nodes, note that while submissions to the normal_N and normal_W queues may work for awhile longer, you are encouraged to change your scripts to *not* specify those queue names. Instead use the "model=neh" or "model=wes" attributes described above.

If you have been running on Harpertown nodes and want to continue to do so, please add the "model=har" attribute to your resource request. At this time, no other changes have been made to the PBS queues.

Here are some examples of requests that ask for resources sufficient for a 128-process job:

  #PBS -l select=16:ncpus=8:model=har
  # to run all 8 cores on each of 16 Harpertown nodes

  #PBS -l select=32:ncpus=4:model=har
  # to run on only 4 cores on each of 32 Harpertown nodes
  # (note: will be charged for 32 nodes = 256 cores)

  #PBS -l select=16:ncpus=8:model=neh
  # to run all 8 cores on each of 16 Nehalem nodes

  #PBS -l select=11:ncpus=12:model=wes
  # to run all 12 cores on each of 11 Westmere nodes
  # (4 cores in 11th node will go unused)
  								

For further assistance or questions, please contact NAS User Services.

NAS Operations
NAS User Services
support@nas.nasa.gov
800.331.8737
650.604.4444