Scheduling Supercomputer Jobs under Existing Power Consumption Constraints

• Published in: Lobachevskii journal of mathematics Vol. 45; no. 10; pp. 5082 - 5091
• Main Authors: Kiselev, E. A., Yarovoy, A. V., Telegin, P. N., Baranov, A. V.
• Format: Journal Article
• Language: English
• Published: Moscow Pleiades Publishing 01.10.2024; Springer Nature B.V
• Subjects: Algebra; Analysis; Computation; Energy consumption; Fault tolerance; Geometry; Mathematical Logic and Foundations; Mathematics; Mathematics and Statistics; Probability Theory and Stochastic Processes; Scheduling; supercomputer; HPC; workload manager; job management system; job scheduling; power consumption
• ISSN: 1995-0802; 1818-9962
• DOI: 10.1134/S199508022460612X

The growth of energy consumption is increasingly becoming the subject of research by scientists. The increase in the load on power grids negatively impacts the fault tolerance and reliability of computing. This is especially true for the IT infrastructure of actively developing regions. Due to the significant increase in energy consumption, it is necessary to pay special attention to its uninterrupted operation. In this paper, we considered various approaches to ensuring the energy stability of a computing system using the functionality of various batch systems. We proposed a method for scheduling user jobs under conditions of constantly changing threshold energy consumption. The proposed solution, in comparison with other scheduling methods, allows to significantly reduce the number of jobs discarded for execution due to lack of power. The results of the simulation confirm the efficiency of the proposed solution.