- Verwendung hierarchischer Monitoringdaten zur Berechnung job-spezifischer Energieverbräuche
Littmann, Thomas; Müller, Matthias S. (Thesis advisor); Lankes, Stefan (Thesis advisor); Wassermann, Christian (Consultant)
Aachen : RWTH Aachen University (2025)
Bachelor Thesis
Bachelorarbeit, RWTH Aachen University, 2025
Abstract
Nowadays, there is a great awareness for sustainability and energy efficiency in computing. With compute clusters drawing up to several megawatts of power, there is a need to attribute the energy consumption of the cluster to the jobs scheduled to run thereon. This way cluster operators can bill the users for the incurred energy consumption and carbon emissions of the computation jobs. Previous research has investigated the power consumption of processes and developed models for estimating a process’s power consumption. However, these models and software tools focus on the estimation while a process is running. With the availability of performance metrics of a compute cluster, a job’s energy consumption can be estimated after its end using additional metrics from the environment of the cluster and the compute node hardware. This thesis proposes and implements two models providing energy consumption estimates for the compute nodes and three weighting factors attributing the energy consumption to individual jobs running on the nodes. The models use metrics from the power distribution units (PDU), power supply units (PSU) and hardware sensors from the monitoring data gathered during the cluster operation for the estimates. The validation shows a deviation between model estimates and PSU energy consumption on average of 0.08% and 1.93% for the models. The proposed weighting factors differ in their estimates and each encourage different user behaviour or leave energy unattributed.
Institutions
- Department of Computer Science [120000]
- Chair of High Performance Computing (Computer Science 12) [123010]