Practical Experience Using a Computational Model for the Design of Heterogeneous Distributed Software

Heterogeneous cluster environments are becoming an increasing popular platform for executing parallel applications. Efficient heterogeneous programs must account for the differences inherent in such an environment. We propose the HBSP model of computation as a framework for developing applications for heterogeneous clusters of workstations. The utility of the model is demonstrated through the design and analysis of the scatter and one-to-all broadcast algorithms. Extensive experimentation illustrates the benefits of using the model for heterogeneous program development. By hiding the non-uniformity of the underlying system, the HBSP model provides a framework that embraces the heterogeneity of the underlying system.