SKA SDP Benchmark Monitor - benchmon

SKA SDP Benchmark Monitor, also known as benchmon, is a tool for monitoring resource usage, energy consumption, and tracing callgraphs. It is designed to analyze and optimize the performance and energy efficiency of applications, whether running on a single node or across multiple nodes. It is non-intrusive and runs in the background, allowing applications to execute without interference from the monitoring process.

This guide describes the main features, installation, usage, and visualization capabilities of benchmon.

• Resource usage monitoring: benchmon can monitor and record a wide range of system metrics, including:

  • CPU: Average and per-core usage, usage per space, and frequencies.

  • Memory & Swap: Used, cached, and free memory and swap.

  • Network & Infiniband: Bandwidth.

  • Disk: Bandwidth and IOPS.

• Energy consumption

  • RAPL measurements: Uses perf for power profiling and energy consumption.

  • Wattmeter integration: Supports Grid5000 wattmeters and BMCs for node-level power monitoring.

• Callgraph tracing: benchmon uses perf to capture callgraphs for plot annotation and to generate flame graphs for in-depth profiling.

Getting started

• Installation
• Basic procedure
• Monitoring options and flags
• SW/HW contexts
• Pre-defined benchmarking levels

Tutorials

• Mono-node example
• Multi-node example
• Self-Calibration Pipeline
• Pre-defined level benchmark

Developer guide

• benchmon package
• benchmon.common package
• benchmon.common.slurm package
• benchmon.exceptions package
• benchmon.hardware package
• benchmon.hardware.advanced package
• benchmon.hardware.gatherers package
• benchmon.run package
• benchmon.software package
• benchmon.software.gatherers package
• benchmon.visualization package