Availability of green energy controls IT

The electricity demand of IT infrastructures is continuously increasing, thereby increasingly harming the environment. To reduce their CO2 footprint, IT systems are therefore more frequently integrated into so-called microgrids that provide direct access to electricity from renewable energy sources. However, the local availability of solar and wind energy varies greatly. Researchers at the Berlin Institute for the Foundation of Learning and Data (BIFOLD) at TU Berlin have developed a new approach for IT systems that only allows flexible workloads, such as training tasks for machine learning systems, to be processed when they can exclusively be powered by renewable energy sources. Their publication "Cucumber: Renewable-Aware Admission Control for Delay-Tolerant Cloud and Edge Workloads" will be presented at Euro-Par 2022.

The demand for computing power is increasing not only in industry but also in research year after year, driving the energy consumption of distributed computing systems and the associated carbon emissions higher and higher. "Data centers are already responsible for more than one percent of the world's energy consumption, and this number is expected to rise further — not least as the Internet of Things (IoT) as well as edge and fog computing continue to gain ground," explains BIFOLD Fellow Prof. Odej Kao, Professor of Distributed and Operating Systems at TU Berlin. One approach for more sustainable and cost-saving cloud and edge computing systems is the direct equipping of IT infrastructure with renewable energy sources such as sun or wind. However, especially smaller data centers are not always able to consume all the generated electricity, leading to energy surpluses at times and shortages of renewable energy at others. A problem that can only be partially alleviated through energy storage or balancing consumption within locally confined power grids.

Better planning of flexible workloads

Flexible workloads that tolerate some delay in execution are common in cloud environments but can also occur in time-critical edge computing environments. For example, an automated, intelligent traffic control system must constantly adapt to new situations by iteratively improving local machine learning models based on new data. The exact timing and extent of such training tasks, however, are subject to a certain degree of flexibility.

To utilize excess renewable energy at computing nodes more effectively and thus reduce CO2 emissions and electricity costs, the BIFOLD researchers propose a new approach: Flexible workloads, such as training tasks, can be rejected by the computing system if there is not enough energy from renewable sources available. Their system predicts the free capacity of a computing system as well as the expected energy consumption and energy production. This way, it can be determined whether additional tasks can be added to the system's queue without consuming energy from the public power grid. Using probabilistic forecasts, the system can be individually configured: either it accepts only workloads that are almost guaranteed to run on green energy or a potential use of grid power is not fully excluded if more tasks can be processed overall. "The increased use of renewable energies requires flexibility in electricity consumption, which is difficult to coordinate in highly distributed and heterogeneous systems," says Philipp Wiesner, a doctoral student at TU Berlin. "We see our approach as an integral component of a decentralized data processing system that optimizes the use of renewable energy through local decisions."

Publication:
Philipp Wiesner, Dominik Scheinert, Thorsten Wittkopp, Lauritz Thamsen, Odej Kao, Proceedings of the 28th International European Conference on Parallel and Distributed Computing (Euro-Par): "Cucumber: Renewable-Aware Admission Control for Delay-Tolerant Cloud and Edge Workloads".
https://arxiv.org/pdf/2205.02895.pdf

Further information is gladly provided by:
Philipp Wiesner
TU Berlin
Department of Distributed and Operating Systems (DOS)
Tel.: 0049 (0)30 314-26260
Email: wiesner@tu-berlin.de