The paper “Data Center Performance Model for Evaluating Load Dependent Energy Efficiency” is now available in the proceedings from the International Conference on ICT for Sustainability (ICT4S) at Atlantic Press:http://2016.ict4s.org/program/accepted-papers-for-ict4s-2016/

The authors, Daniel Schlitt, OFFIS,
and Wolfgang Nebel, C.v.O. University of Oldenburg, propose the novel Load Dependent Energy Efficiency metric (LDEE) to enable data center operators to optimize their facilities and thereby decrease operational expenses. In the EU project M2DC, the LDEE will be used to determine the energy efficiency of IT systems in order to dynamically deploy applications in the most energy efficient way. In addition to this, the extension of LDEE to alternative compute nodes like ARM processors, GPUs or FPGAs will be analyzed.

© Atlantis Press. This article is distributed under the terms of the Creative Commons Attribution 3.0 Unported License, which permits commercial and non-commercial use, distribution and reproduction in any medium, provided the original work is properly cited.

Abstract — Energy efficiency metrics are important tools for data center operators to optimize their facilities and thereby decreasing operational expenses while strengthen competitiveness. However, commonly used metrics like Power Usage Effectiveness do not consider productivity or suitable proxy indicators, thus lacking the ability for correctly comparing energy efficiency between data centers. Also, other known metrics which consider productivity, do this in a subjective way, i.e. results are only comparable for the same definitions. In order to address these shortcomings we proposed the Load Dependent Energy Efficiency (LDEE) metric, which uses a combination of utilization, performance, and power models to provide detailed efficiency data. By using load dependent models the concrete workloads are abstracted realizing comparability. Furthermore, models are trained with public information such as hardware specifications and benchmark results to avoid disruption of operation and thereby increasing applicability. This paper focuses on the utilization and performance models of LDEE.

Index Terms — Data center, performance modeling, load modeling, energy efficiency metric, benchmarks