Dynamic scheduling in heterogeneous multiprocessor architectures

Abstract

A MPAHA (Model for Parallel Algorithms on Heterogeneous Architectures) model that allows predicting parallel application performance running over heterogeneous architectures is presented. MPAHA considers the heterogeneity of processors and communications. From the results obtained with the MPAHA model, the AMTHA (Automatic Mapping Task on Heterogeneous Architectures) algorithm for task-to-processors assignment is presented and its implementation is analyzed. DCS_AMTHA, a dynamic scheduling strategy for multiple applications on heterogeneous multiprocessor architectures, is defined and experimental results focusing on global efficiency are presented. Finally, current lines of research related with model extensions for clusters of multicores are mentioned.