Research On The Energy Storage System In Heterogeneous Environment

The idle power consumption of a server can reach 60% of the peak-power. The ratio of throughput and power of the server under light workload is lower than the one under heavy workload, in other words,most energy is wasted while workload of the server is light. In order to save energy, high ratio of throughput and power is expected. One common approach is shutting down a portion of servers while the total workload is light, so that the utilization of remaining servers will raises and power consumption will be effective. As there must be a complete data set in remaining servers, the amount of servers that can be closed is limited.A power-efficient heterogeneous storage system is proposed. The servers named light-server which can have a high ratio of throughput and power under light workload were imported to the data center. The original servers which have higher peak-power but is power inefficient under light workload is called heavy-server. An energy saving scheduling policy that use light-server to handle the light workload is implemented. The policy has two stages. The first stage is that while the workload is light, only light-servers are start-up and can service. As the workload raises higher, heavy-serves becomes available for data server, which is second stage. A new data layout is designed for energy saving scheduling policy. In the new data layout, one replication of data is distributed in fixed servers and different replications lay overlapping in all servers, so that more servers can be shut down while workload is light and the restructure of data can be parallel.We evaluate our power-efficient heterogeneous storage system against the common one. Our power-efficient heterogeneous storage system can reduce 40% power consumption. In addition, the power-efficient heterogeneous storage system combines with the Intel running average power limit can obtain better energy effect.