STUDY OF PARALLEL SOLUTION ORGANIZATION FOR EXTERNAL AERODYNAMICS PROBLEMS BASED ON SPLITTING SCHEMES

Abstract

The aim of this work is to study the ways to organize parallel solutions of external aerodynamics problems. A hybrid parallel-conveyor method for numerical solution of two-dimensional problems is considered. It allows to simulate the flow of viscous compressible fluids around objects of complex shape. A parabolized system of Navier-Stokes equations is considered, for the numerical solution a finite-difference algorithm is chosen. Due to its features (cost-effectiveness and stability in the study of boundary layers of moving bodies) this algorithm was preferred. To implement a nonlinear finite-difference scheme, the internal iterations are used in each main section. The developed parallel algorithm consists constructively of nested iterative loops. The system of equations is solved at each internal iteration. It is organized in two stages. At the first stage the equations of motion are solved; at the second stage the density is determined. At each fractional step of the internal iteration, one-dimensional data arrays are calculated. The paper uses the method of splitting the operator by physical processes. For the numerical solution of the problem, the factorization of the stabilizing operator is carried out. The scheme of the organization of the process of problem solving is given in each internal iteration. The paper proposes the principle of organizing parallel computing. The internal parallelism of the physical problem is used here. To implement the parallel algorithm, a computing environment is specially selected. It contains a decisive field of computing devices connected by switching connections, each of computing device has its own RAM. Besides computing environment contains a control device. The parallel algorithm uses a communication topology between worker processors. Reducing the dimension of the problem (to 2d) allows to save time on data exchange between the processors. In this paper, time estimates of the effectiveness of the developed parallel algorithm for each internal iteration are carried out. The use of the parallel run method and the proposed principle of organizing parallel calculations allow to increase the effectiveness of solving problems of such class.