awesomebytes / roscpp_rospy_benchmark

roscpp vs python benchmarks

Tests done on a: Intel® Core™ i7-6700HQ CPU @ 2.60GHz × 8

16GB RAM, GeForce GTX 1070/PCIe/SSE2 Ubuntu 18.04.5 ROS Melodic

Rate of publication

Checking rate with rostopic hz.

Simple message (std_msgs/String)`

Max rate of publication

Publisher Python: 33513~ Hz

Publisher C++: 50505~ Hz

Publisher Python ROSCPP: 45000~ Hz

Big message (sensor_msgs/Image)

Image size: 512x512

Max rate of publication

Publisher Python: 1870~ Hz

Publisher C++: 1600~ Hz

(Weird?)

CPU usage & Memory usage

Publisher

Simple message (std_msgs/String)

10 Hz

Python: 1% CPU, 43MB RAM C++: 0.3% CPU, 8.9MB RAM

30 Hz

Python: 2.3% CPU, 43MB RAM C++: 1% CPU, 8.9MB RAM

50 Hz

Python: 3.3% CPU, 43MB RAM C++: 1.7% CPU, 8.9MB RAM

100 Hz

Python: 6.7% CPU, 43MB RAM C++: 3.0% CPU, 8.9MB RAM

Python ROSCPP: 5.6% CPU, 47.7MB RAM

Complex message (sensor_msgs/Image)

10 Hz

Python: 2% CPU, 71MB RAM C++: 1%CPU, MB RAM

30 Hz

Python: 6% CPU, 71MB RAM C++: 3.3%CPU, 24MB RAM

50 Hz

Python: 10% CPU, 71MB RAM C++: 5.3%CPU, 24MB RAM

100 Hz

Python: 19% CPU, 71MB RAM C++: 10.0%CPU, 24MB RAM

Delay of messages

Using rostopic delay.

Images only

Python: 0.001s C++: 0.001s

Subscribers

Subscribing to std_msgs/String

CPU usage

30Hz

Python: 1% C++: 1%

50Hz

Python: 1.3% C++: 1.3%

100Hz

Python: 2.7% C++: 2.7%

1000Hz

Python: 14% C++: 16%

(Weird)

Current conclusions:

For tiny messages:

Having publishers in C++ may be a bit more efficient.

Having subscribers, not really.

For big messages:

If the data needs to be processed, it will for sure be faster in C++.