Research On Load-Derivation-Forecasting-Based Facing Network Servers Energy Control

Recently, along with the real-time tasks in place, the customers’requirement on QoS is higher and higher. People while enjoy enormous benefit that the information system brings, but one thing was almost entirely ignored----the energy consumption issue. At the same time, as China’s power failures serious degree of deepening, coupled with the economic crisis put forward higher request to each enterprise operators of reduce operating costs, how to improve the server utilization and reduce server energy cost became the focus of the industrial circle and academic circles to consider. The research of this paper has significance to the reasonable assignment of energy saving control, and increasing the utilization ratio of resource.At present, stream media tasks make up a larger percentage of network tasks, so this paper analyzes the stream media server increasing energy consumption problems, main tasks include:1. Based on the known results, this dissertation firstly study the Influence of hard disk on server’s performance, and imporved the method of combining Linux/proc file system with loading kernel modules (LKM), realized the purpose of acquires the four types of network resources(Computing Resource, hard disk Resource, Traffic Resource and Memory Resource). So, this method can comprehensive describe the servers load.2. About the forecast for the network resources, this paper proposes the methods of weight-load-derivation-forecasting-based. This method is based on periodicity of stream media tasks, use the stability of curve rate variation, improve the method of earlier calculation method, aimed at the date fractal similarity, not the graphics. Afterward, smooth the data, and ensure the forecast accuracy.3. About the energy conservation, this paper proposes the methods of the schedule policy based on prediction of low load. The strategy use the results of forecast can reasonable scheduling server’s hibernate quantity and time, transfer the tasks, put the server into hibernation, and wake up before the peak load arrive, won’t cause greater influence to the service performance, so as to achieve the purpose of energy conservation.4. This paper sets up the experiment platform of streaming media application, implements and verifies the method of weight-load-derivation-forecasting-based, and the feasibility of energy saving policy.