Analytical Multi-core Performance Models for Parallel Applications
Application speedup modeling, parallel applications, analytical models
Application’s speedup models were, most of the time, limited by Amdahl’s law. After that, Gustafson introduced the concept of fixed time approach that brings new horizons on parallel applications speedup study. Since then, many other studies have emerged showing other aspects of parallel speedups. This study aims to model the negative effect on speedups of increasing the number of cores caused by the decrease of memory performance. Two models are proposed to capture this behavior. The results of experiments demonstrate that the proposed models are able to model this behaviour with very good approximation even when the models are obtained by a fraction of measurements points, using less resources in the modeling task.