Actions

NewProjectChecklist

• What precisely will be your contribution?
  • We will investigate the effectiveness of improving the scalability of AMG for hyperbolic (time dependent) problems by taking multiple timesteps.
  • We attack the problem from a practical results oriented aproach, instead of just improving MG.
• What question are you answering?
  • Parallelization is hard for coarse girds in AMG.
  • Can the parallelization problems of AMG be fixed by taking multiple timesteps as extra work?
  • How can we best use the extra time in AMG solves?
  • How can we best load balnance AMG computations.
• Why should anyone care?
  • All engineering problems, duh.
• Who would be interested in this research?
  • MG people, engineering community,
  • ??? parallelization (proof that you should consider the bigger problem, don't focus on just the system)

• What larger question does this address?
  • Techniques for parallelizing MG
• What work is similar to what you've suggested?
  • Techniques in survey paper, do more work in MG to eat up idle time.
  • literature on method of lines solutions to differential equations.
• What previous work do you build on? What do you provide a superior alternative to?
  • build on method of lines computation
  • build on other's analysis of parallel performance in AMG.
• How is your result different from and better than this prior work?
  • Instead of making MG faster, we're going to increase the timestep of the hyperbolic solve.
• What journal are you targeting?
  • --: Find a journal we'd like to publish in.

• What, precisely and in detail, is your new result?
  • Accuracy vs scalability analysis, benefit of our approach. (on petascale machines)
• What standard should be used to evaluate your claim?
• What concrete evidence shows that your result satisfies your claim?
  • Performance analysis graphs, side by side comparison to other approaches.
  • execution time vs processors
  • processor utilization vs time on alot of processors
  • strong scalability
  • weak scalability
• What new knowledge have you contributed that the reader can use elsewhere?
  • New approach for parallelization. Don't focus on speeding up MG, focus on speeding it up where it's used.
  • --: need analogy here.

Evaluation chart

• project idea
  • multiple timesteps in one MG solve
• dependent theory
  • NA, MG, parallelzation techniques
• how well explored is the topic. Is it state of the art?
  • relatively new+open, this is a new direction, state of the art.
• difficulty
  • for community
  • for you
• research versus development
• theoretical or application driven.
• impact - high medium low
• prerequisites - dependencies amongst your categories.
• human resources

• Summarize your project idea in 30 seconds? Read it aloud and time yourself.

• focus on lessons learned
  • identify a general problem,
  • find a specific problem
  • reading the agency's request to decide what we want to go into
  • implement the method to find out where the problems are
    • Dr. Olson is doing this soon
    • read code to existing methods
  • do the writing while planning/doing the research
    • answered mary shaw questions
  • give examples/analogies in your presentation
    • motivating cardiac stuff
    • diagrams of our idea
• focus on bring them up to speed
  • cardiac problems
  • multigrid process
  • parallelization approach
  • scalability
  • ?? load balancing

• multigrid: use
• multigrid: how it works
• step 1: parallelization for efficency
• hitch: multi-grid <> parallel
  • list how it kind of works
• state problem with it clearly
  • possibly give analogy
• other's solutions (brief, summerize)
• our approach
  • do two multigrid solves?
• existing technique in ODEs
  • show how two timesteps can increase complexity
• show high level example of accuracy vs time to run a simulation
• compare
• our strategy
  • give initial implementation
  • paper

• what conference do we want to hit
• how can we use our resources
  • Dr. Olson doing this, lots of good knowledge in AMG
  • Dr. Kale has lots of parallel computing, consultant
  • Rob has math knowledge of domain, good knowledge of a problem domain this could apply in
  • Chao has lots of hands on experience optimizing parallel codes
• how will we evaluate our success?
  • performance vs efficiency vs scalabilty vs accuracy
• answers to mary shaw questions?
• hand out core document?

• Motivation for time dependent problems
  • cardiac
  • use this to motivate the importance of linear solvers ->
• show how n is increase
• hot new linear solver -> multigrid
  • state it's advantages quickly
  • show how it works
  • explain advantages
• graphic: small problem vs bigger problems (motivation for parallelization)
• parallelization of multigrid, problems
  • explain

Ideas for the second application

Evaluation chart

• project idea
• dependent theory
• how well explored is the topic.
• difficulty
  • for community
  • for you
• research versus development
• impact - hihg medium low
• prerequisites - dependencies amongst your categories.

Things to think about

• human resources
• focus focus focus.
• reserach impact vs. necessity for project development.
• is it state of the art?

• 2-3 slides
• 15 minutes
• one slide on how you've improved.