Monitoring¶

NVIDIA FLARE provides system monitoring capabilities to track job and system lifecycle metrics for federated learning jobs. This feature enables comprehensive system-level monitoring and visualization of FLARE operations.

Overview¶

The monitoring system focuses on system-level metrics rather than training metrics, providing insights into:

Job lifecycle events
System performance metrics
Resource utilization
Network statistics

This differs from machine learning experiment tracking by focusing on system-level metrics rather than training metrics.

Key Components¶

The monitoring system leverages the following components:

StatsD Exporter: Collects and exports system metrics
Prometheus: Scrapes and stores the metrics
Grafana: Visualizes the collected metrics

These components work together to provide a complete monitoring solution:

FLARE job and system events are captured
StatsD Exporter processes these events
Prometheus scrapes the metrics from StatsD Exporter
Grafana provides visualization and dashboards

Configuration¶

For detailed configuration instructions and setup steps, please refer to the Monitoring README.

The configuration involves:

Setting up StatsD Exporter
Configuring Prometheus
Installing and configuring Grafana
Setting up FLARE monitoring components

Visualization¶

Grafana dashboards provide visual representations of the collected metrics, including:

Job status and progress
System resource usage
Network statistics
Component health

For detailed information about available dashboards and metrics, please refer to the Monitoring README.

Summary¶

The monitoring system provides:

Real-time visibility into FLARE operations
System-level metrics tracking
Comprehensive visualization capabilities
Integration with industry-standard monitoring tools

For more detailed information about setup, configuration, and usage, please refer to the Monitoring README.