PARALLEL FLOW CALCULATIONS ON TRANSPUTERS

Abstract

An explicit finite element scheme based on a two step Taylor-Galerkin algorithm allows the solution of the Euler and Navier-Stokes equations on unstructured grids. Calculations on unstructured grids are inherently parallel and very time consuming so that speeding up the runtime is the main problem which had to be solved. We have implemented an OCCAM algorithm which solves fluid flow calculations on a network of 32 T800 transputers, connected as a chain. To get the desired linear speedup we implemented an efficient load balancing strategy for the network that keeps all processors working at disjoint tasks. A first approach was the division of the grid elements, not the nodes. The resulting speedup of 27 shows how even the total work has been distributed. Additionally an on-line visualization of the computational results on a graphic screen is provided. So, in contrast to a picture drawn by a plotter after the whole computation we get a graphical image of the evolving flow during the runtime of the algorithm.