A tool based on traffic traces and stochastic monotonicity to analyze data centers and their energy consumption

Abstract

We present a tool to study the trade-off between energy consumption and performance evaluation. The tool uses real traffic traces to model arrivals, and it allows to consider general discrete arrival processes. Some servers are switched on (resp. off) if the monitored QoS becomes less (resp. more) than the {\it up} (resp. {\it down}) threshold. A set of threshold couples and the cost function taking into account both the performance measure and the energy consumption are provided by the user. The tool determines the best one for this cost function among the analyzed scenarios. Our method is numerically based but it takes into account some stochastic properties of the model to speed up the computation.