Quantifying Energy Usage in Data Centers through Instruction-count Overhead

Energy usage of data centers is rising quickly and the electricity cost can no longer be neglected. Most efforts to relieve the increase of energy usage concentrate on improving hardware efficiency, by improving the hardware itself or by turning to server virtualization. Yet, no serious effort is made to reduce electricity usage by targeting the software running in data centers. To be able to effectively target software, a quantification of software overhead is necessary. In this paper, we present a quantification of the sources of overhead in applications that are these days ubiquitous in data centers: web applications. Experiments with three web applications show that up to 90% of the instructions executed to generate web pages are non-essential, in other words overhead, and can be eliminated. Elimination of these non-essential instructions results in an approximately linear decrease in page generation time as well as significantly reduced energy usage. In order to get the rising energy cost of data centers under control it is obligatory to be able to quantify the source of energy cost. In this paper we present an approach how to quantify wasted energy based on a quantification of non-essential instructions that are executed.