Research On Data Center Energy Saving Based On DVFS

With the improvement and promotion of cloud computing technologies,large-scale cloud computing data centers are emerging.In order to ensure the stability of data center operating system and service performance,a large number of servers run at peak power state for a long time.Request load of data center fluctuates with time,some servers resources was under-utilized,resulting in the waste of computing resources.In order to minimize servers' power consumption while ensuring consumer experience and quality of service,this thesis considered DVFS technology and server dynamic load variation,the CPU should work in a low-power state as long as possible,thereby reducing the overall data center power consumption.In this thesis,the entire data center energy-saving scheduling process is divided into three parts.The first part is to get the optimal frequence level.The second part is to decide how servers select the request task in accordance with the scheduling strategy.The third part is how to scale processor frequency.This thesis analysed the composition of the data center power consumption.Analyze the relationship between CPU dynamic power and CPU clock frequency and operating voltage,calculate the processor power consumption in different states,as well as the state transition power consumption.Considering dynamically varing job attributes,a linear combination optimization model is established.However,the task execution time needs to be balanced between the execution performance and the power cost while the reduction of energy consumption would extend.Therefore,this thesis based the total power consumption of the data center as the optimization target,put the dynamic load of the server and the Qo S guarantee of the server as the constraints.To solve this problem,we take three steps.Firstly,select the optimal power consumption corresponding to the clock frequency.Secondly,the distribution strategy is proposed to distribute the request task to the appropriate server.The second algorithm is to design the distribution strategy.The characteristics of the request task are analyzed.Then the task set is transformed into the directed graph according to the arrive time and the amount of computation and the deadline of each request task.This algorithm of finding the shortest path is designed by combining with Dijkstra algorithm,the shortest path is found from the digraph and the shortest path job set is distributed to the corresponding server iteratively.Thirdly,the dynamic scheduling algorithm of power consumption was proposed.In view of the optimization goal of energy consumption model,this thesis attempts to control the dynamic energy consumption of server CPU by dynamically adjusting the clock frequency.Combined with the waiting queue of the processor DVFS technology and the server,a scalable sliding window frequency modulation and voltage regulation algorithm is designed.This thesis presented an On/Off scheduling strategy for waiting queue of job set,select the lowest cost scheduling program,whcih can get the local minimum Energy consumption.The simulations have been carried out based on traces from real date center.Compared with other methods,our algorithm can greatly reduce the total power consumption for cloud data center.