Measuring the Performance of a Power PC Cluster

In this work we present benchmarking results from our Linux cluster of four dual processor Power Macintosh computers with processors G4/450 MHz. These machines are used mainly for solving large scale problems in air pollution and Monte Carlo simulations.We first present some numbers revealing the maximum performance of an individual machine. The results are from the well known LINPACK benchmark and an optimized Mandelbrot set computation.The second set of benchmarking results covers the NAS Parallel Benchmark. These tests are written using MPI and Fortran 77 with some Fortran 90 constructs and are close to the real problems that we solve on the cluster. We also tested the performance of a free implementation of Open MP - the Omni Open MP compiler.The last set of tests demonstrates the efficiency of the platform for parallel quasi-Monte Carlo computations, especially when the vector unit is used for generating the Sobol' sequences.