Storage Analysis and Evaluation of Multiple Supercomputers: Findings, Insights and Suggestions

High performance computers are utilized to address large-scale scientific issues across various fields. Typically, users adopt a wide array of scientific applications and submit jobs to the system. These applications have diverse requirements, including input and output (I/O) operations. After years' usage of the storage system, the available capacity, performance and stability of the storage devices are bound to be unable to adapt to the status where the scale of computing data and the production speed continue to increase. It is paramount to understanding the storage system's usage, pinpointing possible bottlenecks, and guiding optimization or upgrade techniques. We collect and analyze the monitoring data from 7 supercomputers dedicated to computational fluid dynamics (CFD) applications. After thorough analysis and evaluation, we propose some practical methods, including continuous monitoring, capacity expansion and device upgrade, as well as the construction of a centralized storage pool.