LS-DYNA Observations

• Parallel performance sensitive to conflicts from other jobs

• Elapsed time relative to single process T90 run:

  • 2 processors (dedicated) runs ~20% longer
  • 4 processors (dedicated MSP) runs ~5% longer
  • 8 processors (dedicated MSP) approximately equal run times

• Elapsed times within 5% with 16 times the memory

  • Less waiting in the batch queue for memory resources
  • Ability to analyze larger models

• Recommend running with 2 or 4 processes dedicated.2 processes MP_HOLDTIME=large value4 processes NCPUS=4 /etc/cpu -a 1 lsdyna I=input …

Previous slide

Next slide

Back to first slide

View graphic version