Energy-saving Scheduling Of Servers For Cloud Data Center

The rapid popularity of cloud computing services has led to rapid growth in the number of data centers.Currently,the data center make a large number of idle servers exist in order to meet certain system performance,ensure high resource utilization and make some range of services,which causing a huge energy consumption waste and incresing the overall operating costs.Reducing the energy consumption of servers in data center can effectively reduce the overall energy cost,dispatching the energy consumption of the server from the energy-saving point of view reasonably to minimize the server operation when we make sure the service quality is not degraded.At present,resource scheduling in the data center focuses on the improvement of resource utilization rate,while there are low energy utilizaiotn rate and high energy cost problems.In this paper,idle servers are sheduled to a low-power sleep state to reduce energy waste.The research on energy saving and consumption of clusters composed of heterogeneous servers and clusters of homogeneous servers is mainly done as follow:In the case of server single-sleep mode energy consumption scheduling,we study how to reduce the energy consumption of the server.The single-sleep power consumption scheduling model of the server is established by analyzing the power consumption characteristics of different states and the energy consumption of state transition.Aiming at the optimization goal of energy consumption model,a sliding window selection scheduling algorithm is designed by using the idea of ”arranging as many servers as possible to sleep when satisfy the user's request”.The real data center server running log is used to simulate the experiment.Compared with the currently used energy scheduling policies AlwaysOn,Reactive and SoftReactive.The experimental results show that the sliding window selection scheduling algorithm reduces the power consumption of the scheduling cycle by about 10% when the average response time is increased by about40% compared with the AlwaysOn strategy without sleep scheduling.This paper analyzes the power consumption of server clusters in active state,sleep state and transition state,and establishes a multi-sleep mode energy consumption scheduling model for servers.The energy consumption scheduling problem is transformed into an integer linear programming problem,and the problem is reduced by using the divideand-conquer thought,and the BackUpdate algorithm is proposed to solve the ”breakpoint”problem of the sub-scheduling cycle.Getting the optimal solution of the server energy consumption scheduling.In the server state definition,by introducing ACPI-defined computer sleep state,different sleep power consumption and conversion costs vary.Because of the large size of the multi-sleep energy consumption scheduling problem,by dividing the scheduling time into several sub-scheduling cycles with the same length,each subscheduling cycle uses the linear optimization engine Cplex to solve the optimal solution.For the ”breakpoint” problem in the sub-scheduling cycle,the BackUpdate algorithm is proposed to update the decision variables to ensure the correctness of the solution.Under the MATLAB simulation,compared with the energy consumption scheduling strategy commonly used in the data center,the BackUpdate algorithm can significantly reduce the energy consumption of the data center.