changing rmw of the auto-generated ros2 node leads to increased CPU consumption

Hello, I run a node on Ubuntu 20.04 (ros2 humble) with a standard rmw (fastrtps), look in htop and see 10-15% CPU load by my node, stop the node, change the rmw on cyclondds, start it and see the CPU load 35-40%...I tried to rebuild the node with the selected cyclondds in the Matlab ros settings - it didn’t help

Can someone explain how to fix it?

[Walter Roberson]

What happens if you stop the node, do not change the rmw, and then start the node again? It would be interesting to see whether it is the act of changing the rmw that causes the problem or the act of starting the node again.

Likewise it would be interesting to see what happens if you change the rmw to the same value it already is (to test whether the act of writing to the rmw triggers the problem.) And it woud be interesting if you were to change rmw back to its original value after running the second time, to test whether the problem is related to the range of the new value you changed it to.

[Josh Chen]

Hi Dmitriy,

For Simulink code generation perspective, we generate exactly same code no matter wich rmw_implementation been selected.

I wonder if this behavior is caused by difference between fastrtps and cyclonedds by nature? Looks like people are seeing different CPU usage comparing these two in the past.

-Josh