… Make Sure You Measure Something Meaningful

When measuring energy consumption, there is also the question of which program paths your code executes. As Freitag points out, errors are not allowed to occur during the measurements. Errors often cause an operation to abort and the program to generate less load.

You also need to measure under real-life conditions if possible. Developers sometimes get in a hurry and simplify the test environment in ways that could affect the measurement results. During their measurements, the oCIS developers resorted to the simplest authentication method (Basic Authentication). As a result, the system repeatedly decoded the same security token in a computationally intensive way. After switching to OpenID Connect, the entire system was measurably faster.

Energy measurements need to cover as many of the software parts as possible. Freitag recommends identifying several practical use cases and testing them. With file sharing software, it makes a difference whether a company shares only a few very large files or an extremely large number of small files. You need to take an average value into account in your measurements.

You can read the results of the SLE study online if you are interested [16], and you will find the test programs on GitHub [3]. OCIS developers are hoping to disclose their comparisons of ownCloud, oCIS, and Nextcloud. At the time this article was written, an internal validation was still in progress. But you can measure the performance yourself with the open source cdperf tool.

Conclusions

The C, Pascal, and Go languages help developers achieve an optimal compromise between runtime, memory usage, and energy efficiency – at least that's what the SLE study suggests. Nevertheless, the Go compiler does not automatically produce energy-efficient applications. Too many other factors play a role.

According to Freitag, it is primarily the software architecture that influences the performance and, indirectly, the energy efficiency of an application. For example, communication among Internet services can consume more energy than saving a file. In addition, efficient implementation is important, and that – in turn – depends on the use case.

There is no doubt that it is worthwhile to get started with resource-efficient programming now. See the box entitled "Tips for Energy-Efficient Implementation" for some tips on greener coding.