Architecture

Architectural Features:

• Platform independent infrastructure for common operations across facilities
• Scalable - Demonstrated in production over 4 years up to 27K nodes.
• Integrated streaming and historical analysis
• Strategy for long term storage and recall of data for trends

Components:

Data Collection:

• Lightweight Distributed Metric Service (LDMS) - data collection and transport. Run-time tunable fidelity (including subsecond) with no adverse impact on application performance
• Collectors for application and platform data and integration of external collection (e.g. kokkos)

Analysis:

• Streaming analysis and active feedback capabilities
• Integrated System Console (ISC) - analytics database, retrieval and display
• LogDiver+Baler - event tagging and relationship discovery
• A variety of analysis tools including ML/AI based tools for identification

Actionable Intelligence:

• Assessments of System Conditions, Application Resource Utilization, and Performance provide actionable insights to be used to improve operations, performance, and system designs:
  • Notifications
  • Dashboards and Visualizations

Feedback and Control:

• Transport architecture includes flow of analysis results to humans and the system to enable action as a result of analyzed conditions. Examples include:
  • Application rebalancing based on utilization imbalances
  • System scheduler decisions based on congestion areas or application profiles
  • System administrator decision support