IDED (formerly known as SDIA) completed the first end-to-end environmental impact assessment of a server-side software application. Are we at the start of a fundamental change?

The energy measurement was successfully conducted solely using software-based methods — the IPMI and Intel RAPL interfaces. This is an important step towards our own power resolver method. Although it seems like a small step, in conjunction with previously developed formulas — such as the Digital Resource Resolver — it represents significant progress in attributing energy consumption to individual applications based on consumed digital resources.

Additionally, with the help of the Boavizta API and the Boavizta agent, the embedded environmental impact of an actual server can now be determined — including lifecycle data (embedded CO₂ emissions) and transparent energy consumption. The integration of real-time data from Electricity Maps finally allows for the conversion of energy consumption into CO₂ emissions based on the actual national power mix.

A significant observation: Intel RAPL underestimates energy consumption by 26%

An interesting insight: The Intel RAPL driver captures about 26% less than the actual energy consumption on this system — as it does not measure the disks and other peripheral devices. Additionally, the open compute servers used show a significant difference between idle state (60 W, CPU load < 5%) and full load (300 W, CPU load 100%). 60 W at idle is still a lot, but it's progress.

Next step: Attribution

The next major step is to attribute the server's environmental impact — including energy — to a specific IT application running on the same server. This attribution is crucial, as an isolated assessment of the entire IT infrastructure footprint can lead to misconceptions.

Infrastructure exists to operate IT applications — therefore, the infrastructure impact must be attributed to the application, along with unused resources reserved by the application. For sustainable IT, users, developers, and operations must come together to ask the tough questions: Does this application need to run constantly? Do we need it at all? How can we add new features without increasing the environmental impact?

Next steps and community

The following initiatives are underway and need support:

• More test servers: Those who can provide servers in a data center for measurements are warmly invited to join the DEF community.

• Hackathons: In-person events to further develop the measurement methodology are planned.

• Public Testbed: The current test setup will become fully publicly accessible.

• Open Source: The components used will be published as Environmental Data Agent on GitHub.

• Kubernetes Integration: Aknostic and Helio lead the Kubernetes integration — support welcome.

• Open Source Libraries: As part of the SoftAWERE project, open source libraries are examined for energy consumption and environmental impact.

A heartfelt thanks to everyone who made this milestone possible — especially Benoit Petit (Hubblo/Boavizta), Christoph (Helio), Jeroen & Joost (Blockheating), John Laban (Open Compute Foundation), Jurg & Flavia van Vliet, Maarten Kamoen (Aknostic) as well as Fridtjof Chwoyka and Ludger Ackermann (Data Center Excellence). And of course, to the researchers: Jens Gröger and Felix Behrens (Öko Institut), Stefan Naumann (Environmental Campus Birkenfeld), Patricia Lago and Ivano Malavolta (VU Amsterdam) and many others.

This work is supported by the German Environment Agency, the Federal Ministry for Economic Affairs and Climate Action (BMWK), and the European Commission.