ZED Benchmark

This package contains a set of benchmarks to test the performances of the ZED ROS2 Wrapper

Usage

$ ros2 run zed_topic_benchmark zed_topic_benchmark --ros-args -p topic_name:=<name_of_the_topic_to_monitor>

The benchmark stops and prints a report when a duration or sample-count limit is reached, or when interrupted with Ctrl+C. For example, benchmark 1000 messages and save the report to a file:

$ ros2 run zed_topic_benchmark zed_topic_benchmark --ros-args \
    -p topic_name:=<name_of_the_topic_to_monitor> \
    -p test_sample_count:=1000 \
    -p log_file_path:=/tmp/benchmark_report.txt

Multi-camera launch

The zed_test_ipc.launch.py launch file starts a multi-camera setup and one benchmark node per camera (composed in the camera container, or as separate processes):

$ ros2 launch zed_topic_benchmark zed_test_ipc.launch.py \
    cam_names:=[zed_front,zed_back] cam_models:=[zed2i,zed2] cam_serials:=[12345678,87654321]

Launch arguments:

• cam_names, cam_models, cam_serials: arrays describing the cameras to start.
• disable_tf: disable TF broadcasting for all cameras. [Default: False]
• use_ipc: load the benchmark nodes as components in the camera container, with intra-process comms enabled, instead of starting them as separate processes. [Default: True] The benchmark nodes are given subscription_mode:=typed, which is what makes the intra-process path usable: see Measuring Intra Process Communication.
• subscription_mode: subscription path used by the benchmark nodes: auto, generic or typed. Applied to both branches so that use_ipc:=True and use_ipc:=False are directly comparable. [Default: typed]
• qos_reliability: subscriber QoS reliability, best_effort or reliable. [Default: best_effort] Both work on ZED topics, which are published Reliable. A Reliable subscriber only fails against a Best Effort publisher, so keep the default when benchmarking a topic from some other node whose reliability you have not checked.
• qos_durability: subscriber QoS durability, volatile or transient_local. [Default: volatile]
• qos_history: subscriber QoS history, keep_last or keep_all. [Default: keep_last]
• qos_depth: subscriber QoS depth, used by keep_last. [Default: 1]
• avg_win_size: window size of the running averages. [Default: 500] Applied to both branches: the report min/max are tracked on the windowed average, so an asymmetric window would make the two use_ipc runs incomparable.
• topic_name: topic to benchmark, without the /zed_multi/<cam_name> prefix. [Default: /point_cloud/cloud_registered]
• test_duration_sec: benchmark duration in seconds. 0.0 runs until Ctrl+C. [Default: 0.0]
• test_sample_count: number of messages to acquire before stopping. 0 runs until Ctrl+C. [Default: 0]
• log_file_path: file where the report is saved (the camera name is appended to keep per-camera reports separate). Empty disables file logging. [Default: ""]

Note: with use_ipc:=True all benchmark nodes share the camera container, so a finite test_duration_sec/test_sample_count will stop the whole container once the limit is reached. Keep the limits at their defaults and stop with Ctrl+C to benchmark composed nodes for an arbitrary time.