This section describes a recommended workflow to identify performance issues in the executed graph. This workflow may change as more analytics plugins are added. However, the fundamental principle should not change, as the goal is to maximize the throughput of the graph in a streaming case, and provide the best scaling performance with respect to the serial run.
The Flow Graph Analyzer provides the following capabilities for analyzing flow graph performance:
- Display the graph for which the execution trace is captured. See the Preferences section for details on how to enable loading .graphml files that contain graphs with cross-graph edges.
- Display the trace information and highlight parallel performance issues.
- Map poorly scaling time regions to nodes executing at that time.
- Compute the critical path of the graph.
- View compute statistics for the computational nodes based on the execution traces.
- View prioritized diagnostics.
Follow the steps in this section to analyze performance.