J-OPT: A Joint Host and Network Optimization Algorithm for Energy-Efficient Workflow Scheduling in Cloud Data Centers

Abstract

Workflows are a popular application model used for representing scientific as well as commercial applications, and cloud data centers are increasingly used in the execution of workflow applications. Existing approaches to energy-efficient workflow scheduling in cloud computing environments have primarily focused on the optimization of server utilization. The majority of works have ignored the impact of scheduling decisions on the data center network (DCN). However, studies have revealed that the DCN consumes 10-20% of the total data center power, and this percentage could rise much higher depending on the utilization level of the data center. This paper proposes an energy-efficient workflow scheduling approach (J-OPT) that jointly optimizes the power consumption of servers and networking elements in cloud data centers. J-OPT considers precedence constraints and data dependencies among workflow tasks as well as communication requirements among task instances in the formulation of topology-aware scheduling decisions. The proposed approach is evaluated using synthetic and real world workflow traces in a simulated environment. Results of the experiments demonstrate that J-OPT outperforms state-of-the-art algorithms in terms of total power savings by 8% and 30% under high and low data center utilization levels, respectively.