Computational issues

1
Computational issues

• Use FSUs 512 processor IBM 690p server
• Third fastest university owned supercomputer in
  the US
• Science-aware parallelization
• Predict regions likely to experience short
  time-scale phenomena and concentrate
  computational resources there
• Avoid fine granularity where possible
• Use Monte Carlo techniques for rare-event
  simulation when required, to avoid fine
  granularity
• Efficiently parallelizable through replication
• Faster versions of traditional parallelization
  techniques
• Stochastic versions of traditional domain
  decomposition techniques
• Trade computation for communication
• Mixed shared and distributed memory
  parallelization
• Optimize sequential component too
• Cache-aware computation

Issues
Solutions

• Large time scale
• Small system size
• Fine grained parallelization
• High communication cost
• Adaptive computations
• Regions experiencing short time-scale phenomena
  simulated with a finer resolution
• Spatial decomposition and granularity change
  dynamically, and quickly, with time
• Need fast and efficient load balancing strategies

2
Preliminary resultsNanocomposite simulation

• Model
• Matrix-nanotube interface modeled with springs
• An extra force term computed for atoms attached
  to springs
• Springs can break, requiring substantial increase
  in computations in that region

• Experimental parameters
• Nanotube with 1000 atoms
• Spring probability 0.05
• Probability of a spring breaking in an iteration
  0.01
• Load increase factor due to spring break 200
• Disturbance region depth 3
• Number of time steps 100

Spring
Polymer matrix
3
Experimental results