Big Simulation Needs

[bcaddy]

Here's some notes I took at ATPESC on needs for big simulations. I hope this can be a discussion/knowledge store for future simulations that the group does.

Dealing with Data

• Smart layout of directories to uniquely sort data
• Tooling (scripts) to help me find the files I want
• Compress data either losslessy or lossy
• "version" data by including log files, run information, etc with the data
• Automate tape storage
• Pegasus - workflow pipeline generator for this kind of stuff

Planning

• Extra testing at scale to determine correctness
• Pathfinder runs: smaller runs to verify the larger simulation will work. Can also serve a resolution study. Can also create mock ups of features you expect to see if the amount of resources required are reasonable.
• Look for optimizations based on science rather than pure computation.
• Have inflight diagnostics to stop the run if something goes wrong.
  • Check for conservation, are fields positive like they're supposed to be? etc