Middlewares are complex. Middleware comparisons are complex. There are many dimensions of tradeoffs that go into the design and implementation of middlewares, so any evaluation needs to consider many different use cases. The goal of this repo is to have a bunch of little programs to experiment with various middlewares to learn more about their pros and cons.
- Raw TCP (to provide a baseline)
- Raw UDP (to provide a baseline)
- ROS 1
- MQTT/RabbitMQ
- Zenoh
- ZeroMQ ("pure", what others use)
- ZeroMQ (
ign_transportversion) - Native DDS
- Fast DDS 2.0 (used to be known as Fast RTPS)
- Eclipse Cyclone DDS
- Connext
- DDS via ROS 2
- Fast DDS 2.0
- Eclipse Cyclone DDS
- Connext
| Event Name | Description | Robot Application | Evaluation Metric |
|---|---|---|---|
| Hello world | Send strings from A to B. | Nav targets | Pass/fail |
| 1KB pub/sub | Send 1 KB messages at 1 kHz | Arm control | Latency and jitter |
| 1MB pub/sub | Send 1 MB messages at 30 Hz | Raw image stream | Latency and jitter |
| 10MB pub/sub | Send 10 MB messages at 10 Hz | Raw point clouds | Latency and jitter |
| Good WiFi pub/sub | Various messages over ideal WiFi | Mobile robots | Latency and jitter |
| Bad WiFi pub/sub | Various messages over intermittent WiFi | Mobile robots | Latency, jitter, drop rate, reconnect speed |
| Discovery | Discover various (large) numbers of topics | Startup | Discovery time |
Refer to the README.md file in the directory for each contestant for how to set up and run the events for that contestant.
It can be a pain to set up a "bad" network interntionally, but we can instead use tc to fake it:
Add 10ms of "fake" latency:
sudo tc qdisc add dev lo root netem delay 10ms
Remove the "fake" latency:
sudo tc qdisc del dev lo root netem
When using mininet, Avahi can be doing random talking which increases the packet count unnecessarily.
To stop Avahi on Ubuntu:
sudo service avahi-daemon stop