Performance evaluation of DECK combining multithreading and communication on Myrinet and SCI clusters

This paper presents a performance evaluation of DECK (Distributed Execution and Communication Kernel), a multithreaded parallel programming environment for clusters of SMPs, with the parallel implementation of the classical Mandelbrot fractal generation and Laplace’s Equation algorithms. The applications have been run on Myrinet and SCI clusters and the results are compared to corresponding MPI implementations. The comparison shows that DECK is able to achieve very good performance when multithreading is combined with communication, without the need of multiple processes on a single machine.