Webinar 2020 Visualizing job usage on the Compute Canada systems with the ViewClust Python package

From SHARCNETHelp
Jump to navigationJump to search

Compute Canada account usage via the Slurm scheduler is a dominant factor that is used the determine the priority of each job that enters the queue. The priority of each job may have substantial influence on the wait time in the queue. Understanding the usage pattern of an account on the Compute Canada systems can help establish expectations for throughput and also help determine if alternative job submission strategies could result in shorter wait times, shorter time to results, and/or higher throughput. The "sacct" command in Slurm provides information required for assessing an account's usage on the system, but the tabular data can be difficult to summarize and use to generate the relevant usage measures that are used to calculate job priority by the scheduler. ViewClust is a Python package being developed by SHARCNET staff to summarize and interactively visualize job usage data. This general interest webinar describes the summary figures and measures made available by the ViewClust package and provides a demonstration of its usage on Compute Canada systems.