Optimal Speed Setting for Cloud Servers With Mixed Applications

Abstract

The technique of workload dependent dynamic power management can dynamically and flexibly adjust power and speed according to the current workload. It has been well recognized that improving server performance and reducing energy consumption can be achieved by employing the technique of workload dependent dynamic power management. It is an effective way to deal with the power and performance tradeoff for cloud servers. In this study, applications are divided into different classes, which have different characteristics. The server speed is different in processing tasks from different types. Hence, we explore the technique of variable and task type dependent server speed management to optimize the server performance and to minimize the power consumption of a server with mixed applications. This is also a kind of workload-dependent dynamic power and speed management to deal with the power and performance tradeoff. We establish an M/G/1 queueing model for a server with variable and task type dependent speed, so that our investigation can be conducted analytically. We formulate the problems of power constrained performance optimization and performance constrained power minimization as multivariable optimization problems, and solve the problems by efficient numerical algorithms. We provide numerical data to compare the performance of a server with the optimal speed setting to that of a server with a constant speed, and to compare the power of a server with the optimal speed setting to that of a server with a constant speed. It is shown that the reduction in the average response time can be as high as 9.9% and the reduction in the average power consumption can be as high as 8.0%.