Methodology for predicting the energy consumption of SPMD application on virtualized environments

Abstract

Over the last decade, the computing clusters have been updated in order to satisfy the increasing demand of greater computational power for running applications. However, this increasing is transformed in more system en- ergy consumption, which results in financial, environmental and in some cases with social consequences. Hence, the ideal is to achieve an scenario that allows the system admin- istrator to find a trade-off between time and energy-efficiency for parallel algorithms on virtualized environments. The main objective of this work is based on developing an analytical model to predict the energy consumption and energy delay product (EDP) for SPMD applications on virtual environments. The SPMD applications selected are designed through a message passing interface (MPI) library with high communication volumes, which can generate im- balance issues that affect seriously the execution time and also the energy-efficiency. Our method is composed by four phases (characterization, tile distribution model, mapping and scheduling). This method has been validated using scientific applications and we observe that the minimum Energy and EDP values are located close to the values calculated with our analytical model with an error rate between 4% and 9%.