Research and Development of Evaluation Tools on User Job Level Index of HPC Cluster
Yiqin Gao, Yang Zhou, Rui Li, Jianwen Wei, Yichao Wang, James Lin · 2025
With the construction of supercomputing network and the widespread application of supercomputing, the number of supercomputer users is growing rapidly.However, users from different disciplines exhibit varying levels of proficiency in using supercomputers.To effectively enhance user capabilities and increase the efficiency of cluster resource utilization, this paper proposes a quantifiable evaluation system for user job level on HPC clusters.We take the supercomputer platform of Shanghai Jiao Tong University as an example, to introduce the design and implementation of the evaluation system, including key aspects such as indicator selection, data processing, weighting, and index calculation.The selected indicators reflect the frequency and efficiency of users' job utilization of cluster resources and their parallel computing capabilities.We processed the collected raw data using log-transformation and normalization methods, then we employed the entropy weight method and analytic hierarchy process for indicator weighting, ultimately calculating the user job level index.We have developed an opensource evaluation tool for the system, which has been applied on the real platform and has demonstrated the effectiveness of the evaluation system.The tool helps HPC administrators intuitively compare the job levels of different users, identify and guide those with lower job levels, and monitor changes in the overall job level of the cluster.