H. Chi and E.L. Jones (USA)
Parallel Computing, Parallel Streams, Monte Carlo, LCG, Spectral Test
Monte Carlo simulations are thought to be very easy to par allelize; however, the quality of these parallel Monte Carlo computations depends greatly on the quality of the parallel random number generators used. Much of the work dealing with parallel random streams has been aimed at partitions of one generator that uses a fixed modulus. In contrast with this approach, we propose generating parallel streams us ing separate Linear Congruential Generators (LCGs) with different moduli. When implemented for parallel computa tion, correlation between streams is critical. With the help of spectral test, we show that the analysis of the correlation between any two streams can be carried out in a number theoretic framework. The proposed scheme for generating parallel random number streams is especially good for het erogeneous and unreliable computing environments.
Important Links:
Go Back
