A DATA PARTITION METHOD FOR MEMS-BASED STORAGE DEVICES IN A DISTRIBUTED COMPUTING ENVIRONMENT

This paper aims to minimize power consumption and sustain customer satisfaction for a datacenter. In some distributed computing environments, especially in a cloud computing environment, the most concerned issue for a datacenter designer is power consumption instead of its speed. This is because the storage devices always consume much electricity. To save power, we can assign to each device proper workload and turn off some spare devices. Therefore, partitioning data and allocating them among the storage devices becomes the first priority. However, due to some complex constraints, the data partition problem is very time-consuming or even NP-hard. Therefore, we propose a gradient-based method to deal with the problem. The novelty of the proposed method lies in that the discretized combinatorial problem is solved by a calculus-based technique. Experimental results show that significant energy is saved by shutting down as many uncalled-for devices as possible. The results also suggest that other similar optimization problems can be solved by the linearly-convergent method.