A two-level parallelization algorithm for the direct simulation Monte Carlo method of problems of rarefied gas dynamics

A two-level parallelization algorithm is proposed for the direct simulation Monte Carlo (DSMC) method. The algorithm is based on the decomposition of the computational domain into subdomains, which in turn are divided into blocks. The algorithm is implemented in the DSMC software code using MPI and OpenMP technologies. The algorithm was tested on the problem of spherical gas expansion into a vacuum from an evaporating surface, which is of interest for astrophysical applications. The problem is characterized both by large gradients of physical parameters and large time of stationary solution achievement. The calculations were carried out on the resources of the Polytechnic – RSK Tornado supercomputer. When solving the problem, OpenMP technology is used to calculate blocks of subdomain by single process associated with single node, each containing 28 threads with shared memory. MPI technology is used to exchange data on migrating between subdomains particles among nodes with distributed memory. The speedup of the algorithm was 35 on 64 nodes of the supercomputer. Additionally, a single-level decomposition algorithm was tested, implemented only by MPI tools both for calculating subdomains within them. The high efficiency of the single-level algorithm for a relatively small number of threads is shown. The boundary of the preferred use of the two-level algorithm is defined.