Tips for Writing a Successful NWSC Proposal

General submission format

Five-page request typically

• Project information (title, lead, etc.)

• Project overview and strategic linkages

• Science objectives

• Computational experiments and resource requirements (HPC, DAV, and storage)

Supporting information

• Multi-year plan (if applicable)

• Data management plan

• Accomplishment report

• References and additional figures

Remember your audience: Computational geoscientists from national labs, universities and NCAR

• Don’t assume they are experts in your specialty

Be sure to articulate relevance and linkages between:

• Funding award,

• Computing project,

• Eligibility criteria, and

• NWSC science objectives (as appropriate)

Don’t submit a science proposal

• Panel is not reviewing the science, they are reviewing the computational need so describe the science in detail sufficient to justify the computational experiment's proposed.

Most of the request should focus on computational experiments and resource needs

• Effective methodology: Are you using the right computational tools for the job?

• Appropriateness of experiments: Are the proposed experimental configurations necessary and sufficient to answer the scientific question? Are all key experimental parameters justified?

• Efficiency of resource use: Are the resources being used efficiently for the proposed methodologies and experiments?

• The amount of requested resources should be clearly and explicitly calculated based on the justification for the three preceding criteria

  • E.g., University guidelines recommend a table with one row per experimental configuration

Justifying resource needs

HPC

• Cost of runs necessary to carry out experiment, supported by

  benchmark runs or published data

• Yellowstone allocations will be made in “core-hours” (not GAUs).

• Request a small allocation to conduct actual Yellowstone benchmark runs

• Reasonably justified estimates based on runs from other systems will also be accepted.

• Yellowstone core-hours can be calculated as: # of nodes x 16 x job duration (in hours).

DAV—will be allocated, similar to HPC practice

• Simple justification for standard interactive use: # users x 5,000 core-

  hours

• Small allocations will get up to 5,000 core-hours upon request

• Allocation review will focus on larger needs associated with batch use • E.g., for projects conducting GPGPU code development and testing

Glade