Green Configuration: Determining the Influence of Software Configurations on Energy Consumption

• Applicant: Professor Dr.-Ing. Norbert Siegmund
• Subject Area: Software Engineering and Programming Languages
• Term: from 2017 to 2023
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 380077267

Reducing energy consumption of IT systems is fundamentally important for saving cost and reducing CO2 emissions. A largely untapped potential arises from the configuration options a software system provides to adapt it to the application scenario, workload, and underlying hardware. Whereas the first phase of Green Configuration had a focus on deployed configurable software systems (e.g., finding energy-optimal configurations or measuring energy and learning a model for an already developed application), we will shift the focus on the actual development process. The goal of Green Configuration's second phase is to lay the foundations for energy-aware software development. The goal is that energy assessment and optimization become a seamless activity in the development process. Specifically, we aim at forecasting a software configuration's energy consumption for heterogeneous, possibly virtualized environments, mainly based on data collected from existing tests, thereby avoiding the need of heavy-weight measurement setups and lowering the barrier of adoption. At the same time, we aim at supporting practical energy testing of configurable software systems, thereby helping developers in spotting energy regressions and hot spots with their existing testing methodology and development workflows. Both goals are intertwined in that novel methods, tools, and concepts need to be developed to seamlessly integrate the complex and highly sensitive measurement process into the everyday development process. This requires research on how workload, hardware, and environmental changes (e.g., due to staging in CI/CD and virtualization) affect energy consumption in greater detail. Since these variations cause a combinatorial explosion of variations in energy consumption, we will devise novel methods based on probabilistic programming that assess the likelihood and error range of energy consumptions to prioritize test cases. The outcome are uncertainty-aware energy models. By integrating uncertainty as a first-class citizen, deployment, energy improvement, reconfiguration, and additional test runs can well be rationalized.

• DFG Programme: Research Grants
• International Connection: USA
• Co-Investigator: Professor Dr.-Ing. Sven Apel
• Cooperation Partner: Professor Dr.-Ing. Christian Kästner