ros2-comparison

Dora vs ROS2 Benchmark Comparison

Reproducible latency and throughput comparison between Dora and ROS2 using identical workloads.

Prerequisites

• Dora CLI installed (cargo install --path binaries/cli)
• Python Dora package: pip install dora-rs
• numpy and pyarrow: pip install numpy pyarrow
• ROS2 Humble (or later) with rclpy and std_msgs

Methodology

Both frameworks run the same workload:

• Latency: 100 messages per payload size with 5ms sleep between sends
• Throughput: 100 messages per payload size sent back-to-back
• Payload sizes: 0B, 8B, 64B, 512B, 2KB, 4KB, 16KB, 40KB, 400KB, 4MB
• Timestamps: time.perf_counter_ns() embedded in first 8 bytes of each payload
• Metrics: avg, p50, p95, p99, p99.9, min, max latency; messages/sec throughput

Both sides use Python APIs to ensure a fair comparison of framework overhead (not language performance).

Quick Start

cd examples/ros2-comparison
./run_comparison.sh

This runs both benchmarks sequentially and prints a comparison table.

Manual Execution

Dora

BENCH_CSV=results/dora.csv dora run dataflow.yml --uv

ROS2

Terminal 1 (receiver):

BENCH_CSV=results/ros2.csv python3 ros2_receiver.py

Terminal 2 (sender):

python3 ros2_sender.py

Compare Results

python3 analyze.py results/dora.csv results/ros2.csv

Output Format

CSV files contain one row per (mode, payload_size):

latency,<bytes>,<label>,<n>,<avg_ns>,<p50_ns>,<p95_ns>,<p99_ns>,<p999_ns>,<min_ns>,<max_ns>
throughput,<bytes>,<label>,<n>,<msg_per_sec>,<elapsed_ns>,0,0,0,0,0

Files

File	Description
dora_sender.py	Dora benchmark sender node
dora_receiver.py	Dora benchmark receiver node
dataflow.yml	Dora dataflow definition
ros2_sender.py	ROS2 benchmark publisher
ros2_receiver.py	ROS2 benchmark subscriber
run_comparison.sh	Orchestration script
analyze.py	CSV comparison and reporting