Modeling Power Consumption and Saving Potential of a Scalable Multi-server Architecture

C.-H. Lien, Y.-W. Bai, and M.-B. Lin (Taiwan)

Keywords

Cluster Computing, Resource Allocation, Modelling and Simulation, DelayPower Product.

Abstract

In current computing systems, multiple servers are often used to enhance a system’s performance. However, multiple servers seriously increase power consumption. Thus, reducing power consumption is critical issue in such system designs. This paper first proposes a queueing model for a scalable multi-server architecture in order to estimate the reduction of the power consumption. In addition, a stochastic service request process is used to model the operational characteristics of the proposed multi-servers. Due to the trade off between computer performance and power consumption, a scalable power management method is proposed to balance and minimize the “Delay-Power Product” as a comparative index in a multi-server system. Given a certain set of physical parameters, the initial results show that the optimal number of allocated servers can be between two and three, and the “Delay-Power Product” of our scalable power management policy will be better compared to other policies when the average arrival rate is less than a certain service rate.

Important Links:



Go Back