Macro Level Models of Power Consumption for Servers in Distributed Systems

Abstract

In order to realize green eco-society, the total electric power consumption of information systems has to be reduced. In this chapter, we discuss novel models of the power consumption of a server in distributed systems. We first classify distributed applications of information systems into three types: computation (CP), communication (CM), and storage (ST) types of applications. We first measure at macro level the power consumption of a whole server where the types of applications are performed. We do not consider how much electric power each hardware component like CPU consumes at micro level. The power consumption of a server depends on not only hardware component but also software component. Based on the experimentations, we imply power consumption models of a server to perform types CP, CM, and ST of application processes at macro level. Then, we propose algorithms to select a server in a set of servers so that not only QoS (Quality of Service) requirement like response time and deadline constraints hold but also the total power consumption of the servers is reduced based on the power consumption models. We evaluate the selection algorithms compared with traditional algorithms like round robin algorithms in terms of the total power consumption and satisfiability of QoS requirements.