Parallel computations of turbulent transonic flows on a meiko computing surface

Abstract

The chapter focuses on a parallel version of an explicit finite-volume scheme for turbulent transonic flows and the details of its implementation on a T800 Meiko Computing Surface. The code employs a cell-vertex-storage arrangement of flow variables with a Lax–Wendroff time-marching scheme, which advances the solution to the steady state. The chapter discusses the approaches to deal with the extensive data dependencies arising from the discretization approach along with related implementation issues. Parallel speed-up and efficiency levels achieved with a file-decomposition strategy are reported for inviscid and turbulent transonic flow over a two-dimensional aerofoil. Efficiency values are in the range of 70–99%, depending on problem type, mesh size, and granularity of the domain decomposition. One particular problem encountered was the load imbalance across the processor array, tending to limit the maximum possible code performance. The mesh used for examining the speed-up and efficiencies was rather coarse and tended to exaggerate this effect. Future developments of the program include the addition of a multi-grid routine, an essential measure to counteract the very slow convergence rates of the time-stepping algorithm.