Always use_sim_time?

[doganulus]

I currently investigate the timing aspects of Autoware, which is critical for testing and simulation but also its overall safety. Normally it's desirable to build real-time systems on a linear, increasing, and abstract time model, which helps the determinism and stability of the system.

One of ROS2's design articles talks about the issue of clock and time and mentions that such time abstraction can be achieved via publishing to the /clock topic and setting use_sim_time parameter at launch. Now the parameter name use_sim_time may imply that it is only for simulation. But, as far as I can see, it can be used to have better control over timestamps even at deployment time and provide a greater parity with simulation. Yet I am not familiar enough with network issues or other performance considerations.

So I wonder what arguments can be made against setting use_sim_time always and controlling timestamps via an explicit clock publisher in Autoware?