Abstract | ||
---|---|---|
Energy-based billing as well as energy-efficient software require accurate knowledge of energy consumption. Model-based energy accounting and external measurement hardware are the main methods to obtain energy data, but cost and the need for frequent recalibration have impeded their large-scale adoption. Running Average Power Limit (RAPL) by Intel enables non-intrusive, off-the-shelf energy monitoring, but only on a per-socket level. To enable apportioning of energy to individual applications we present E-Team, a non-intrusive, scheduler-based, easy-to-use energy-accounting mechanism. By leveraging RAPL, our method can be used on any Intel system built after 2011 without the need for external infrastructure, application modification, or model calibration. E-Team allows starting and stopping measurements at arbitrary points in time while maintaining a low performance overhead. E-Team provides high accuracy, compared to external instrumentation, with an error of less than 3.5%. |
Year | Venue | Field |
---|---|---|
2017 | 2017 USENIX ANNUAL TECHNICAL CONFERENCE (USENIX ATC '17) | Industrial engineering,Computer science,Parallel computing,Energy accounting,Multi-core processor,Cost accounting |
DocType | Citations | PageRank |
Conference | 1 | 0.35 |
References | Authors | |
24 | 6 |
Name | Order | Citations | PageRank |
---|---|---|---|
Till Smejkal | 1 | 4 | 1.75 |
Marcus Hähnel | 2 | 110 | 6.54 |
Thomas Ilsche | 3 | 172 | 14.92 |
Michael Roitzsch | 4 | 58 | 7.80 |
Wolfgang E. Nagel | 5 | 1800 | 167.93 |
Hermann Härtig | 6 | 686 | 64.40 |