Concurrent DSMC Method Using Dynamic Domain Decomposition

Abstract. In the current study, a parallel two-dimensional direct simulation Monte Carlo method is reported, which incorporates a multi-level graph-partitioning technique to dynamically decompose the computational domain. The current DSMC method is implemented on an unstructured mesh using particle ray-tracing technique, which takes the advantages of the cell connectivity information. Standard Message Passage Interface (MPI) is used to communicate data between processors. In addition, different strategies applying the Stop at Rise (SAR) [7] scheme is utilized to determine when to adapt the workload distribution among processors. Corresponding analysis of parallel performance is reported using the results of a high-speed driven cavity flow on IBM-SP2 parallel machines (memory-distributed, CPU 160 MHz, RAM 256 MB each) up to 64 processors. Small, medium and large problems, based on the number of particles and cells, are simulated. Results, applying SAR scheme every two time steps, show that parallel efficiency is 57%, 90 % and 107 % for small, medium and large problems, respectively, at 64 processors. In general, benefits of applying SAR scheme at larger periods decrease gradually with increasing problem size. Detailed time analysis shows that degree of imbalance levels off very rapidly at a relatively low value (30%~40%) with increasing number of processors applying dynamic load balancing, while it, at a value of 5~6 times larger, increases with increasing number of processors without dynamic load balancing. At the end, the completed code is applied to compute a near-continuum gas flow to demonstrate its superior computational capability