How much power does your server consume? Estimating wall socket power using RAPL measurements

Full system electricity intake from the wall socket is important for understanding and budgeting the power consumption of large scale data centers. Measuring full system power, however, requires extra instrumentation with external physical devices, which is not only cumbersome, but also expensive and time consuming. To tackle this problem, in this paper, we propose to model wall socket power from processor package power obtained from the running average power limit (RAPL) interface, which is available on the latest Intel processors. Our experimental results demonstrate a strong correlation between RAPL package power and wall socket power consumption. Based on the observations, we propose an empirical power model to predict the full system power. We verify the model using multiple synthetic benchmarks (Stress-ng, STREAM), high energy physics benchmark (ParFullCMS), and non-trivial application benchmarks (Parsec). Experimental results show that the prediction model achieves good accuracy, which is maximum 5.6 % error rate.

[1]  Thomas Ilsche,et al.  An Energy Efficiency Feature Survey of the Intel Haswell Processor , 2015, 2015 IEEE International Parallel and Distributed Processing Symposium Workshop.

[2]  Dhabaleswar K. Panda,et al.  Evaluation of Energy Characteristics of MPI Communication Primitives with RAPL , 2013, 2013 IEEE International Symposium on Parallel & Distributed Processing, Workshops and Phd Forum.

[3]  Wu-chun Feng,et al.  Towards energy-proportional computing for enterprise-class server workloads , 2013, ICPE '13.

[4]  OrgerieAnne-Cecile,et al.  A survey on techniques for improving the energy efficiency of large-scale distributed systems , 2014 .

[5]  Lizy Kurian John,et al.  Complete System Power Estimation Using Processor Performance Events , 2012, IEEE Transactions on Computers.

[6]  Rafael Mayo,et al.  Reducing the cost of power monitoring with DC wattmeters , 2014, Computer Science - Research and Development.

[7]  Wolfgang E. Nagel,et al.  Power measurement techniques on standard compute nodes: A quantitative comparison , 2013, 2013 IEEE International Symposium on Performance Analysis of Systems and Software (ISPASS).

[8]  Rahul Khanna,et al.  RAPL: Memory power estimation and capping , 2010, 2010 ACM/IEEE International Symposium on Low-Power Electronics and Design (ISLPED).

[9]  Christos Kozyrakis,et al.  Full-System Power Analysis and Modeling for Server Environments , 2006 .

[10]  Kai Li,et al.  The PARSEC benchmark suite: Characterization and architectural implications , 2008, 2008 International Conference on Parallel Architectures and Compilation Techniques (PACT).

[11]  Hermann Härtig,et al.  Measuring energy consumption for short code paths using RAPL , 2012, PERV.

[12]  A. Dell'Acqua,et al.  Geant4 - A simulation toolkit , 2003 .

[13]  Song Huang,et al.  Measurement and characterization of Haswell power and energy consumption , 2015, E2SC '15.

[14]  Bruce M. Maggs,et al.  Cutting the electric bill for internet-scale systems , 2009, SIGCOMM '09.

[15]  Rajesh Gupta,et al.  Evaluating the effectiveness of model-based power characterization , 2011 .

[16]  Reinaldo A. Bergamaschi,et al.  Empirical and analytical approaches for web server power modeling , 2014, Cluster Computing.

[17]  Laurent Lefèvre,et al.  A survey on techniques for improving the energy efficiency of large-scale distributed systems , 2014, ACM Comput. Surv..