PARMONC - A software library for massively parallel stochastic simulation

Abstract

In this paper, the software library PARMONC that was developed for the massively parallel simulation by Monte Carlo method on supercomputers is presented. The "core" of the library is a well tested, fast and reliable long-period parallel random numbers generator. Routines from the PARMONC can be called in the user-supplied programs written in C, C++ or in FORTRAN without explicit usage of MPI instructions. Routines from the PARMONC automatically calculate sample means of interest and the corresponding computation errors. A computational load is automatically distributed among processors in an optimal way. The routines enable resuming the simulation that was previously performed and automatically take into account its results. The PARMONC is implemented on high-performance clusters of the Siberian Supercomputer Center. © 2011 Springer-Verlag Berlin Heidelberg.