Bad network performance with multiple robots in ROS2

[Krzo99]

Hello,

I'm using ROS2 with FastDDS. With one robot (and the PC), the network works well, without any problems.
However, as soon as I add another robot, running the same software as the first, the network grinds to a halt, 90% of the time.
This means, msgs not being received on the PC, or being 10s-15s late. There isn't any topic name collisions, as the same setup works the other 10% of the time.

Each robot has hundreds of topics, publishers & subscribers on ~10 nodes. My theory is, that the network gets overwhelmed with all the multicasting, and grinds the network to a halt, creating a multicast storm. Sometimes the router even crashes/reboots.

Would it be possible to configure fastdds_config.xml file, to disable multicasting, if I know all the IP addresses of the robots & PC.
Could you provide the example, I'm a bit lost,
PCs IP is 192.168.1.123, robot's IP is: 192.168.1.111

Thank you for all the help!

[jonatansw]

Hi Krzo99,

Unfortunatelly not able to help, but we are experiencing the very same thing, especially with images hardly working at all.
Did you make any progress on this?

Thanks

[Krzo99]

We used seperate DOMAIN_ID's for each robot, and connected only important topics with the domain_bridge package. Maybe JesusPoderoso's answer might help you.

[JesusPoderoso]

Hi @Krzo99, @jonatansw, thanks for using Fast DDS.
In deployments where the amount of nodes is really small, the initial peers configuration is recommended.
It would imply adding the IP address of the participants to which the communication wants to be established, per participant.
The Discovery Server mechanism would also reduce significantly the network traffic caused by multicast, and it is more scalable than initial peers. In this case, one participant would act as a server (PC) and the other participant(s) (robots) would connect to the server that acts as a hub for meta traffic discovery.
The following link refers to a Discovery Server Vulcanexus tutorial. The Use ROS 2 with Fast-DDS Discovery Server documentation section may be of your interest too, and the C++ and XML configuration is also explained in the Discovery Server section of the Fast DDS documentation.
Hope it helps!

[Krzo99]

In the end, I solved it by using seperate DOMAIN_ID's for every robot, and connecting them using domain_bridge package. It works well, but feels like a hack.