Spatiotemporal Task Scheduling for Heterogeneous Delay-Tolerant Applications in Distributed Green Data Centers

Abstract

A growing number of organizations deploy multiple heterogeneous applications in infrastructures of distributed green data centers (DGDCs) to flexibly provide services to users around the world in a low-cost and high-quality way. The skyrocketing growth in types and number of heterogeneous applications dramatically increases the amount of energy consumed by DGDCs. The spatial and temporal variations in prices of power grid and availability of renewable energy make it highly challenging to minimize the energy cost of DGDC providers by intelligently scheduling arriving tasks of heterogeneous applications among GDCs while meeting their expected delay bound constraints. Unlike existing studies, this paper proposes a spatiotemporal task scheduling (STTS) algorithm to minimize energy cost by cost-effectively scheduling all arriving tasks to meet their delay bound constraints. STTS well investigates spatial and temporal variations in DGDCs. In each time slot, the energy cost minimization problem is formulated as a nonlinear constrained optimization one and addressed with the proposed genetic simulated-annealing-based particle swarm optimization. Trace-driven experiments show that STTS achieves larger throughput and lower energy cost than several typical task scheduling approaches while strictly meeting all tasks’ delay bound constraints. Note to Practitioners —This paper investigates the energy cost minimization problem for a DGDC provider while meeting delay bound constraints for all arriving tasks. Previous scheduling methods do not jointly consider spatial and temporal variations in prices of power grid and availability of renewable energy in DGDCs. Therefore, they fail to adopt such variations to minimize the energy cost of a DGDC provider. In this paper, a new method that avoids disadvantages of previous methods is proposed. It is realized by adopting a hybrid metaheuristic algorithm named GSP to solve a nonlinear constrained optimization problem. Experimental results demonstrate that compared with several typical methods, it reduces energy cost and increases throughput. It can be readily integrated into realistic industrial DGDCs. The future work requires engineers to consider the effect of indeterminacy and uncertainty of green energy on scheduling methods.