[Nsa] RE NSA meeting on supercomputing

[Piotr Smolarkiewicz]

Dear All, 

Please find enclosed the announcement and agenda of the NSA meeting
on supercomputing at NCAR.

Piotr Smolarkiewicz


  NSA meeting on supercomputing, May 30, 9:30am, ML Main Seminar room.
 ----------------------------------------------------------------------

Per NCAR Directorate's request, the NSA will host a meeting to discuss
the Data Center Project, the upcoming June workshop, and the future of
computing at NCAR, in general. The date, time and place are:  May 30,
9:30am, ML Main Seminar room.  


                               A G E N D A

  1. NCAR's response to upcoming OCI solicitations (625 & beyond) 

      *** presentation by Rich Loft

  2. The "Data Center Project" 

      *** presentation by Lawrence Buja and Peter Fox

  3. Compare/contrast NCAR's OCI roadmap and the Data Center Project

      *** arguments by Rich Loft vs. Lawrence Buja and Peter Fox

  4. Massive-cpu/small memory per node architectures appear likely,
      but one size does not fit all.  The challenges of porting our
      codes to a Cray XT3 to achieve useful throughput are presumably
      quite different from the challenges associated with a Blue Gene.
      - What is the range of plausible architectures under consideration
          and the range of impacts/challenges these pose for our
          application groups with regards to operation and memory scaling,
          archival bandwidth and necessity for parallel I/O, etc.?
      - How to make the science staff aware of coming architecture
          changes and the the potential impacts on their research codes?
      - What must be done to secure the additional resources required for
          conversion/optimization of NCAR science codes?

       *** technical issues explained by Rich Loft 

   5. Should NCAR definitely commit to the Data Center idea?
      - Is this too much of a "Faustian bargain" or is this the only way
        to keep supercomputing at NCAR?  Not a trivial question at all.
      
      *** discussion led by Rolando Garcia