Modeling the energy consumption for concurrent executions of parallel tasks
Thomas Rauber, Gudula Rünger · ERef Bayreuth (University of Bayreuth) · 2011
Programming models using parallel tasks provide portable performance and scalability for modular applications on many high-performance systems. This is achieved by the flexibility of a two-level programming structure supporting mixed task and data parallelism. Due to the emerging importance of energy efficiency in high-performance computing, programming models with parallel tasks should be extended to be able to include energy concerns. Based on a well-accepted analytical energy model for a processor's energy consumption, this article explores the energy consumption of parallel tasks with communication that are executed concurrently with other tasks. Simulations show the different energy consumption scenarios for different task cooperations and demonstrate the potential for a flexible energy usage on varying parallel platforms.