1 / 12

Quasi-random Number Generators for Parallel Monte Carlo Algorithms

Quasi-random Number Generators for Parallel Monte Carlo Algorithms. Author: B. C. Bromley Presented by: Shuaiyuan Zhou 10-30-2009. Why & How Sobol’ sequence & subsequences Parallel Algorithm Performance issue. The will and the way.

feleti
Download Presentation

Quasi-random Number Generators for Parallel Monte Carlo Algorithms

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. Quasi-random Number Generators for Parallel Monte Carlo Algorithms Author: B. C. Bromley Presented by: Shuaiyuan Zhou 10-30-2009

  2. Why & How • Sobol’ sequence & subsequences • Parallel Algorithm • Performance issue

  3. The will and the way • Enable parallel systems to take full advantage of the benefits of quasi-random Monte Carlo algorithms • Rapid sequence production. as quick as any pseudorandom generator • Fast convergence rate. much faster than pseudorandom methods • An algorithm of generating quasi-random numbers in parallel systems, using a leapfrog scheme

  4. Main contribution • “A recursion relation which allows an element in a Sobol’ sequence to be quickly calculated from a previous, but not necessarily adjacent, element without determining all of the intervening members of the sequence”. • Enables parallel Monte Carlo algorithms to have each node of a parallel processor step through interleaved subsequences with the same computational load as if it were calculating the original sequence, without any internode communication.

  5. Sobol’ sequence

  6. Recursive computation

  7. Implementing in parallel • In parallel applications, • Breaking up the spatial domain of integration in the M-dimensional cube among processing nodes • Distributing the sequence of sample points among processing nodes • Leapfrog technique: each node skips along the sequence, jumping over those sample points which are handled by other nodes.

  8. Modification to recursion

  9. Algorithm

  10. Algorithm – Cont.

  11. Performance issue • The efficiency of the leapfrog algorithm is highest when the number of nodes is an integral power of 2. • Tests with several parallel supercomputers demonstrate that as many as 106 integration points (up to 6 dimensions) can be generated per second per node in the optimal case. • When the number of nodes is P, the computational load of generating interleaved sequences is, at worst, proportional to the number of nonzero bits in P.

  12. Thank you!

More Related