User-level Workload Analysis for Supercomputers
Qiqi Wang, Yu Shen, Jing Li · 2021
The analysis of workloads is important for understanding how supercomputing systems are used. Gaining insight into user behavior is equally crucial for current system in aspects of job scheduling, promoting allocation efficiency and improving user satisfaction. In this paper, a user-level methodology is proposed to characterize workloads and analyze the interrelation between the extracted characteristics. We apply this methodology to the workloads of two school–level supercomputing system (TC4600, SJTU). We show the distribution of overall jobs characteristics and perform an in-depth analysis of user behavior. Moreover, we present a first investigation into the interior features of ubiquitous BoT (Bag-of-Tasks) in supercomputing system.