zero_copy_rosidl

Overview

This repository is a Proof-of-Concept (PoC) implementation of zero-copy ROSIDL message serialization and deserialization in ROS 2. It serves as a reference implementation for REP-0157, demonstrating the feasibility and performance benefits of zero-copy communication in ROS 2.

Further reads:

• openrobotics/reps#22
• openrobotics/reps#28

Development Environment

Quick Start

Set the ROS domain ID and launch the development container:

export ROS_DOMAIN_ID=<your_domain_id>
./docker/run.sh [--build] [-- COMMAND [ARGS...]]

Use --build to rebuild the image. Commands after -- execute inside the container.

Modifying ROS Core Packages

Modifications are applied during the Docker build when ROS core packages are compiled from source. The script-based approach allows for arbitrary changes beyond simple patches, including removing packages, adding COLCON_IGNORE files, or managing git submodules.

1. Create a shell script docker/files/patches_ros/<package_name>.sh with your modifications
2. For simple patches, reference a corresponding .patch file:

#!/bin/bash
script_location=$(dirname "$(readlink -f "$0")")
script_name=$(basename "$0")
script_name_minus_extension="${script_name%.*}"
git apply ${script_location}/${script_name_minus_extension}.patch

3. For other modifications, implement custom logic directly in the script
4. Make the script executable: chmod +x docker/files/patches_ros/<package_name>.sh

The script executes automatically during the build phase when processing the corresponding ROS core package. The same mechanism applies to external dependencies using docker/files/patches_ext/.

Accessing Running Container

To open a shell in a running container:

docker exec -it <username>_zero_rosidl_devel bash

Replace <username> with your system username (e.g., ekumen_zero_rosidl_devel).

Running Performance Benchmarks

This project includes iRobot's ros2-performance benchmark suite to measure ROS 2 communication performance.

Building the Workspace

The benchmark packages are built automatically during the Docker image build. To rebuild the workspace inside a running container:

# Inside the container, rebuild the workspace
cd $USER_WS
rm -rf build/* install/*
source /opt/ros/jazzy/setup.bash
colcon build

Running the Benchmark

To run the default Sierra Nevada topology benchmark:

# Inside the container, run the benchmark
cd $USER_WS
source $USER_WS/install/setup.bash
cd ./install/irobot_benchmark/lib/irobot_benchmark
./irobot_benchmark topology/sierra_nevada.json

Viewing Benchmark Results

Results are saved in the sierra_nevada_log/ directory:

• latency_total.txt: Summary statistics (mean latency, late messages, lost messages)
• latency_all.txt: Detailed per-message latency data
• metadata.txt: Benchmark configuration parameters
• resources.txt: CPU and memory usage during the test

To view the summary results:

cat $USER_WS/install/irobot_benchmark/lib/irobot_benchmark/sierra_nevada_log/latency_total.txt

Example output:

received_msgs  mean_us   late_msgs late_perc too_late_msgs  too_late_perc  lost_msgs lost_perc
5278           131       0         0         0              0              0         0

Available Topology Configurations

The benchmark includes several predefined topologies in the topology/ directory:

• sierra_nevada.json - Default topology (10 nodes, complex graph)
• mont_blanc.json - Large topology
• cedar.json - Medium complexity
• white_mountain.json - High complexity
• debug_*.json - Smaller topologies for testing

To run a different topology:

./irobot_benchmark topology/<topology_name>.json