Temperature, Power, and Makespan Aware Dependent Task Scheduling for Data Centers

High performance computing data centers are playing increasingly important roles in our daily life. However, as data centers increase in size and number, the power consumption at the data centers has also increased dramatically. We are facing the challenge of reducing energy consumption, lowering down the peak inlet temperature and at the same time meeting short make span requirements. In this paper, we present two dependent task scheduling algorithms to balance the trade-offs among data center´s power consumption, peak inlet temperature, and application´s make span. We compare them with two existing algorithms, i.e., the List algorithm and the Coolest Inlets algorithms. Our extensive simulations show clear advantages of the proposed approaches over the List and the Coolest Inlets algorithms for both homogeneous and heterogeneous data centers.