Scale Ratio Tuning of Group Based Job Scheduling in HPC Systems

Abstract

During the initialization of a supercomputer job, no useful calculations are performed. A high proportion of initialization time results in idle computing resources and less computational efficiency. Certain methods and algorithms combining jobs into groups are used to optimize scheduling of jobs with high initialization proportion. The article considers the influence of the scale ratio setting in algorithm for the job groups formation, on the performance metrics of the workload manager. The study was carried out on the developed by authors Aleabased workload manager model. The model makes it possible to conduct a large number of experiments in reasonable time without losing the accuracy of the simulation. We performed a series of experiments involving various characteristics of the workload. The article represents the results of a study of the scale ratio influence on efficiency metrics for different initialization time proportions and input workflows with varying intensity and homogeneity. The presented results allow the workload managers administrators to set a scale ratio that provides an appropriate balance with contradictory efficiency metrics.