Intelligent Operating System Deployment Strategy Based on KiERA for Optimizing Supercomputers

Supercomputers are essential tools in modern scientific and industrial research, performing complex computations and large-scale data processing swiftly and accurately through high-performance computing (HPC). To maximize the performance of these supercomputers and ensure stable operations, efficient operating system deployment and real-time resource management are crucial. However, current commercial cluster management tools rely on foreign software, leading to increased costs in procurement and maintenance. To address this issue, the Korea Institute of Science and Technology Information (KISTI) implemented its proprietary intelligent cluster management system, KiERA, to enhance the operating system deployment performance for its fifth supercomputer, Nurion. This study focused on adapting the existing proprietary operating system, initially used exclusively for Nurion, to work effectively with KiERA. The performance of KiERA’s operating system deployment was evaluated on 169 compute nodes and compared to the commercial software, Bright Cluster Manager (BCM). Experimental results demonstrated that KiERA completed node deployments in approximately 430 s on average, showcasing improved deployment efficiency over the existing method. Furthermore, the study suggests the potential for deploying up to 13,000 nodes simultaneously in large-scale supercomputer environments. This research contributes to improving supercomputer operational efficiency while reducing dependency on foreign software.