Parallel implementation of Thomas algorithm for the 2D heat equation

Abstract

In this paper was considered a parallel implementation of the Thomas algorithm for the 2D heat equation. MPI was chosen as the technology for parallelization. The numerical solution of the two-dimensional heat conduction problem was solved using the two step iteration process of alternating direction implicit method (ADI). The Thomas algorithm is simple to implement a sequential program, but difficult to parallelize due to dependent data transfers. When applying this method to solve the 2D heat equation, there is a need to implement Thomas algorithm along each direction x-axis and y-axis. It was implemented the parallelization of this problem using the Yanenko method using 1D and 2D data decomposition. In particular, in 2D data decomposition, the Yanenko method was used along each x-axis and y-axis direction. In the article the speedup and efficiency of parallel programs using 1D and 2D data decomposition were shown in the form of tables and graphs. The presented algorithm was tested on a cluster of the computing center of Novosibirsk State University for a different number of points in the computational domain (from 512x512 to 4096x4096). The obtained test results are presented and analyzed, on the basis of which the features of the used decompositions are described.

Keywords

• high-performance computing
• thomas algorithm
• yanenko method
• parallel computing
• adi method
• mpi