Virtual Machine Scheduling Considering Both Computing and Cooling Energy

The problem of high energy consumption of data centres has become increasingly serious. Computing and cooling energy are the most significant parts. On one hand, as the energy consumption of under-utilized resources accounts for a substantial amount of the actual energy draw, load consolidation is widely used for reducing computing energy by increasing resource utilization. On the other hand, to avoid the hot spots, which increase the requirement of cooling capacity and reduce cooling efficiency, load balance is often adopted for better thermal management. These two approaches are contradict with each other. This paper explores the trade off between load consolidation and balance. Our proposed policy Dart Scheduler selects genetic algorithm to generate a near optimum scheduling solution. The results show that, in most cases, our approach can significantly reduce the total energy consumption comparing with the other baseline algorithms. Further, the time cost for Dart Scheduler is tolerable for online deployment and scheduling.