Improved energy-aware stochastic scheduling based on evolutionary algorithms via copula-based modeling of task dependences

Abstract

In this work we apply the copula theory for modeling task dependence in a stochastic scheduling algorithm. Our previous work, as well as the majority of the existing related works, assume independence between the tasks involved, but this is not very realistic in many cases. In this paper we prove that, when task dependence exists, better results can be obtained when it is modeled. Our results show that the performance of the stochastic scheduler is significantly improved if we assume a certain level of task dependence: on average 18% of the energy consumption can be saved compared to the results of the deterministic scheduler, along with 81% of improved test cases, versus 2.44% average savings when task independence is assumed, along with 50% of improved test cases.