Parallel Pseudorandom Number Generation Using Additive Lagged-Fibonacci Recursions

摘要: We study the suitability of additive lagged-Fibonacci pseudorandom number generator for parallel computation. This has a relatively short period with respect to size its seed. However, is more than made up huge full-period cycles it contains. call these different equivalence classes. show how enumerate classes and compute seeds select given class. The use gives an explicit parallelization suitable fully reproducible asynchronous MIMD implementation. To explore such implementation we introduce exponential sum measure quality generators used in serial or parallel. then prove first non-trivial results are aware on this quality.