Sunday, January 15, 2012

Reviewing Computational Projects

Currently i am in the process of reshaping my computational infrastructure. I got from a friend intersting comments about the N body system calculation.

For more details please refer to:

I have finished a framework for vector processing which i am currently evaluating. Some of the results are shown below. The diagram shows the computation time of an N body newton system depending on the number of particles running on an 2x1.8GHz Quadcore Xeon server for different algorithms.

The yellow line represents execution time in case of an completely non parallel implementation. The green line resembles the best theoretical value assuming 8 processors. And the red and blue line are parallel implementations of the algorithm with 8 parallel processes doing the integration using different work distribution mechanisms.

  • The algorithm comes quite near to the theoretical limit (green line) for large number of particles
  • Below 40 particles, the straight forward algorithm is better performing then the parallel algorithms
  • The influence of different work distribution methods on the overall results is rather small.

1 comment:

  1. Updating my site: