Electricity cost minimization in distributed clouds by exploring heterogeneity of cloud resources and user demands

Distributed clouds, consisting of multiple data centers located at different geographical locations, provide a plethora of services to users. They however consume enormous amounts of electricity to power their data centers. The electricity bill is almost 30%–50% of their operational costs. Minimizing the electricity cost of distributed clouds thus is crucial to reduce the operational cost of their cloud service providers. In this paper, we study the problem of minimizing the electricity cost of a distributed cloud, by exploring the heterogeneities of cloud resources and user demands, and time-varying electricity prices, for which we first propose a two-stage optimization framework: dispatching user task requests to different data centers by incorporating the resource demands of the task requests, the workload, and the electricity price in each data center, and energy consumption profiles of different servers in each data center; followed by further energy optimization within each data center through consolidating Virtual Machines (VMs) to different servers to improve the resource utilization ratio. One critical constraint on such task dispatch and VM consolidation is to meet various user Service Level Agreements (SLAs), which include average task scheduling delays and resource demand violation limitations. Under the proposed framework, we then devise efficient scheduling algorithms for task dispatching and VM consolidations, while keeping both the average scheduling delay and resource demand violation limitation of each admitted task met. We finally evaluate the performance of the proposed algorithms through experimental simulations, using real data sets - the real electricity prices and task traces. Experimental simulation results demonstrate that the proposed algorithms are promising.