Profit-Sensitive Spatial Scheduling of Multi-Application Tasks in Distributed Green Clouds

An increasing number of organizations choose distributed green data centers (DGDCs) and use their infrastructure resources to deploy and manage multiple applications that flexibly provide services to users around the world in a cost-effective way. The dramatic growth of tasks makes it highly challenging to maximize the total profit of a DGDC provider in a market, where the revenue, price of power grid, solar radiation, wind speed, the maximum amount of energy, on-site air density, and the number of servers in DGDCs all vary with geographical sites. Different from existing studies, this paper designs a profit-sensitive spatial scheduling (PS3) approach to maximize the total profit of a DGDC provider by smartly scheduling all tasks of multiple applications to meet their response time constraints. PS3 can well utilize such spatial diversity of the above factors. In each time slot, the profit maximization for the DGDC provider is formulated as a constrained nonlinear program and solved by the proposed genetic-simulated-annealing-based particle swarm optimization. Real-life trace-driven simulation experiments demonstrate that PS3 realizes higher total profit and throughput than two typical task scheduling methods. Note to Practitioners-This paper investigates the profit maximization problem for a DGDC provider, while the average response time of all arriving tasks of each application is within their corresponding constraint. Existing task scheduling approaches fail to jointly consider the spatial variations in many factors, including the revenue, price of power grid, solar radiation, wind speed, the maximum amount of energy, on-site air density, and the number of servers in DGDCs. Consequently, they cannot schedule all tasks of multiple applications within their response time constraints in a profit-sensitive way. In this paper, a profit-sensitive spatial scheduling (PS3) method that tackles the drawbacks of previous approaches is presented. It is achieved by adopting a proposed genetic-simulated-annealing-based particle swarm optimization algorithm that solves a constrained nonlinear program. Simulation experiments prove that compared with two typical scheduling approaches, it increases the total profit and throughput. It can be readily realized and incorporated into real-life industrial DGDCs. The future work should improve the proposed method by analyzing the indeterminacy in green energy and the uncertainty in tasks.