Performance prediction methods for evaluating PDE algorithms on MIMD machines
Simon K. Fok, John Ross Wilson, Harry G. Heard, Joseph A Parker · Statistical and Scientific Database Management · 1983
In this paper, a variety of current performance measures for SIMD machines are extended to MIMD machines. The concept of cost-effectiveness is elaborated and used to obtain an aposteriori estimate of the optimal number of processors to be used for a particular problem. Using both probability modeling and simulation, performance prediction methods are developed to evaluate the performance of the Large Eddy Code [7] of NASA-Ames on a tightly-coupled system such as the CRAY X-MP.