Efficient Cluster Parallelization Technology for Aerothermodynamics Problems

Abstract

HPC modeling of gas dynamic and aerodynamic problems is very important for the development of aircrafts, missiles and space vehicles and requires a lot of processor time. For this reason, the numerical codes for such simulations must be efficiently parallelized. This paper presents a technological approach that greatly simplifies the parallelization of problems with unstructured grids. The paper introduces the principle of a unified mathematical address space of the problem for all used cluster nodes. This technology also simplifies grid partitioning. Parallelization of the code is carried out with minimal effort, without changing the main parts of the program. As a result, a single computational code is produced for all regimes – sequential, multi-threaded, and cluster. Performance measurements confirm the good scalability of the method.