Thursday, March 7, 9:30 am in room MTH 3206, University of Maryland,
College Park
Recent Developments in Parallel Pseudorandom Number Generation
Dr. Michael Mascagni
Center for Computing Sciences, I.D.A.,
Bowie, Maryland
mascagni@super.org
Providing high quality pseudorandom numbers for parallel computers supplies
many deep and fascinating mathematical problems. However, one of the more
practical issues in parallel pseudorandom number generation is finding methods
that provide portability and reproducibility across architectures. We
summarize some of the recent developments of our research group in the design
and analysis of pseudorandom number generators for parallel computers that are
portable and reproducible. We will discuss three parallelization techniques:
1. Splitting maximal-period generator's full period into nonoverlapping
subsequences, one for each parallel process.
2. Finding parameterized families of generators and distributing a unique
parameter value for the generator used on each parallel process.
3. Finding generators with statistically similar full-period equivalence
classes and distributing one equivalence class to each parallel process via
seeding.
We present details for each of these techniques applied to certain common and
not so common serial pseudorandom number generators.
|
