A multi-model estimation of distribution algorithm for energy efficient scheduling under cloud computing system

How to manage the applications under computing systems such as a cloud computing system in a more efficient way is a focus problem. The primary performance goal is to reduce the execution time (makespan) of the application. As the need to cloud computing grows, the environmental influence of data centers attracts much attention. This paper aims at the scheduling of the precedence-constrained parallel application to minimize time and energy consumption efficiently. A multi-model estimation of distribution (mEDA) algorithm is adopted to determine both task processing permutation and voltage supply levels (VSLs). Specific operators to decrease execution time and energy consumption are designed. An improvement operator is also designed to enhance the diversity of the non-dominated solutions. The proposed algorithm is compared with the standard heuristic methods and a parallel bi-objective genetic algorithm (bGA). The comparative results show the Pareto solution set by the proposed algorithm is able to dominate a large proportion of those solutions by both the heuristic methods and the bGA.