Skip to content

profiling.md

Latest commit

 

History

History
93 lines (71 loc) · 3.16 KB

profiling.md

File metadata and controls

93 lines (71 loc) · 3.16 KB

Profiling

Profile simulation performance using Linux perf and visualize the results with Hotspot. Uses the profile build mode.

Prerequisites

Profiling requires:

  • Linux perf tool — the perf record command must be available on PATH. On Debian/Ubuntu, install it with sudo apt install linux-tools-common linux-tools-$(uname -r).
  • perf_event_paranoid sysctl — the kernel must allow perf record. Check with cat /proc/sys/kernel/perf_event_paranoid; a value of 1 or lower is needed. To lower it temporarily: sudo sysctl kernel.perf_event_paranoid=1.
  • Hotspot (optional) — required only by open_profile_report() to visualize the profile interactively. Without it you can still use generate_profile_report() to produce the perf.data file and analyze it with other tools.
  • OMNeT++ profile build mode — the project is automatically built in profile mode, which compiles with debug symbols and optimization enabled. See Building for how build modes work.

No extra Python packages are required.

How it works

Both generate_profile_report() and open_profile_report() follow the same pipeline:

  1. The simulation project is built in profile mode (optimized with debug symbols).
  2. The simulation is run under perf record -g --call-graph dwarf, which collects a hardware-level call-graph profile.
  3. The output is written to perf.data (or a custom name via the output_file parameter) in the simulation's working directory.
  4. open_profile_report() additionally launches Hotspot to visualize the profile interactively.

The functions accept all the same filter parameters as run_simulations() (e.g. working_directory_filter, config_filter, run_number, sim_time_limit).

Python API

Generating and visualizing a profile

# Generate a perf.data profile and open it in Hotspot
open_profile_report(simulation_project=inet_project,
                    working_directory_filter="examples/ethernet",
                    config_filter="General",
                    run_number=0,
                    sim_time_limit="10s")

Generating a profile without visualization

report_file = generate_profile_report(
    simulation_project=inet_project,
    working_directory_filter="examples/ethernet",
    config_filter="General",
    run_number=0,
    sim_time_limit="10s")

# report_file is a project-relative path, e.g. "examples/ethernet/lans/perf.data"

The returned path can be analyzed with any perf-compatible tool:

perf report -i examples/ethernet/lans/perf.data

Parameters

Both functions accept:

  • simulation_project — the simulation project (defaults to the current default project)
  • output_file — file name for the perf.data output (default "perf.data")
  • All filter parameters from run_simulations(): working_directory_filter, config_filter, run_number, sim_time_limit, etc.

Command Line

There are no dedicated command-line wrappers for profiling. Use the Python API from within opp_repl.