In this paper we propose an algorithm in which to maximize the number of freed-up machines in data centers, machines that host purely scale-down instances, which are shut down for energy saving at certain maintenance times. We employ an intuitive probability partitioning mechanism to schedule services such that the goal of the maximization can be achieved. Furthermore we propose a set of experiments to test the partitioning rules, which show that the proposed algorithms can dynamically increase the number of freed-up machines substantially.
展开▼
