TY - GEN

T1 - Parallel Random Number Generators Based on Large Order Multiple Recursive Generators

AU - Deng, Lih Yuan

AU - Horng, Jyh-Jen

AU - Tsai, Gwei-Hung

PY - 2009

Y1 - 2009

N2 - Classical random number generators like Linear Congruential Generators (LCG) and Multiple Recursive Generators (MRG) are popular for large-scale simulation studies. To speed up the simulation process, a systematic method is needed to construct and parallelize the random number generators so that they can run simultaneously on several computers or processors. LCGs and MRGs have served as baseline generators for some parallel random number generator (PRNG) constructed in the literature. In this paper, we consider the parallelization problem particularly for a general class of efficient and portable large-order MRGs, in which most coefficients of the recurrence equation are nonzero. With many nonzero terms, such MRGs have an advantage over the MRGs with just a few nonzero terms that they can recover more quickly from a bad initialization. With the special structure imposed on the nonzero coefficients of the new class of generators, the proposed PRNGs can be implemented efficiently. A method of automatic generation of the corresponding MRGs for parallel computation is presented.

AB - Classical random number generators like Linear Congruential Generators (LCG) and Multiple Recursive Generators (MRG) are popular for large-scale simulation studies. To speed up the simulation process, a systematic method is needed to construct and parallelize the random number generators so that they can run simultaneously on several computers or processors. LCGs and MRGs have served as baseline generators for some parallel random number generator (PRNG) constructed in the literature. In this paper, we consider the parallelization problem particularly for a general class of efficient and portable large-order MRGs, in which most coefficients of the recurrence equation are nonzero. With many nonzero terms, such MRGs have an advantage over the MRGs with just a few nonzero terms that they can recover more quickly from a bad initialization. With the special structure imposed on the nonzero coefficients of the new class of generators, the proposed PRNGs can be implemented efficiently. A method of automatic generation of the corresponding MRGs for parallel computation is presented.

U2 - 10.1007/978-3-642-04107-5_17

DO - 10.1007/978-3-642-04107-5_17

M3 - Conference contribution

SP - 289

BT - 8th International Conference on Monte Carlo and Quasi-Monte Carlo Methods in Scientific Computing (MCQMC 08)

PB - Springer-Verlag Berlin Heidelberg

ER -