Generating HPC Job Profiles and Expectations with Time-Series Data - Showcase Presentation
Los Alamos National Lab. (LANL), Los Alamos, NM (United States), Joseph Fullop, USDOE Assistant Secretary for Human Resources and Administration, Brett Leasure Layman · 2020
Summary: Job Profiles and Expectations provide important insights into workloads (Job Profile: window into how a job is running; Job Expectation: Is that job behaving as expected; Provides us with actionable information). Machine learning can be used to group job runs into workload types (Identified groups can then be used for generate expectations); Profiles and Expectations also enable the study of: System-wide events, tracking system changes; System resource utilization and scheduling; Marking log data for further investigation or failures/anomalies.