The Jobstats Job Monitoring Platform for CPU and GPU Clusters
Josko Plazonic, Jonathan D. Halverson · 2024
In 2023, we introduced the Jobstats job monitoring platform [9] which provides user-facing commands and interfaces for inspecting the efficiency of Slurm jobs on CPU and GPU clusters. The platform builds on the Prometheus monitoring framework and the Grafana visualization toolkit. In this poster, we provide updates on the platform, which include additional GPU job metrics, a facility for creating custom job notes, custom output formatting, and improved documentation. Additionally, the tool of the platform that identifies underutilization has been extended to cancel actively running jobs that satisfy certain conditions such as GPU jobs with zero utilization. Jobstats is being used at institutions in Europe and it is starting to see adoption in the U.S.