Power consumption prediction of web services for energy-efficient service selection

Abstract

Listen

Translate article

Service-oriented computing (SOC) is a popular software paradigm that is widely employed in IT industry. SOC uses "services" as the unit of functionality of a software application. The massive wave of SOC applications involves considerable energy consumption of servers, which should not be ignored in large-scale computing environment. When a service requirement can be answered by several web services, the energy consumption for each service to reply to the service request may be different. When this happens, web service selection (WSS) is often required to choose appropriate services to maximize global energy efficiency of SOC applications. Accordingly, this paper proposes a Virtual Power Meter Supported Power Consumption Prediction method for WSS (VPMSPCP). VPMSPCP facilitates choosing appropriate services to minimize wasteful electrical energy from the overall environment of SOC applications. According to our empirical proof, there is a correlation between the power consumption of a service and the status of the server where this service resides. We take advantage of this discovery to develop VPMSPCP by combining a ridge regression model with a well-known web service power modeling method. There are mainly two steps to establish VPMSPCP. First, we develop a virtual power meter (VPM) for each server. VPM is used to estimate the average power of a server under a certain status. Second, we apply the VPM to develop VPMSPCP which estimates power consumption of a web service according to the current status of the corresponding servers. Experiments show that VPMSPCP performs well in improving energy saving in WSS.