Validation of energy profilers
The research is aimed at validating whether so called ‘profilers’ (software tools that can measure the energy consumption of a system or application) are accurate in their readings. We compared energy consumption measurements from the SEFLab setup with the profiler estimations with the test server running specific operations. This allowed us to validate whether profilers can indeed be used to accurately measure the energy consumption of software. The motive for this research was to have validated tooling available for measuring the energy consumption of software.
We expected to come up with a list of profilers accurate enough to give a reliable picture of the energy consumption of a software application. In addition, we expected to indicate the error margin of the profiler estimations.
During this experiment we found that most profilers, even those available commercially, are not suitable yet for measuring the energy consumption of software. The problems started with installing the profilers. In some cases heavy configuration was required which in the end did not provide the desired results, and in other cases it became apparent that the profiler was dependent on extra hardware to perform its tasks. We did find that Joulemeter is accurate for the Windows OS. Though also requiring external hardware (a WattsUp PRO device), Joulemeter readings were fairly accurate compared to the SEFLab readings. For an example, see the figure.
One limitation of this research is that only the CPU has been taken into account. Though we acknowledge that other components play a role in the energy consumption, we did not manage to perform our experiments without influencing these other components. Given that our measurements would be distorted, we decided to exclude the other components from this research.
Partners: Centric, SIG
Period: February 2013 – September 2013
Contact: Erik Jagroep, e.jagroep at centric dot eu