The speed of the algorithm saturates as the number of threads increases
This post was updated on .
I am trying to compute the disparity from a stereo pair. I have two variants of the implementation.
1) Standalone algorithm.
2) The same algorithm integrated into the ROS framework.
The standalone implementation is working 2x faster than the ROS integrated one.
I am using a threaded model of the disparity computation algorithm which uses pthreads.
The ROS integrated algorithm's computation time was not equal to the computation time of the standalone code once the number of threads used for the computation crossed 8.
The other observation is that the ROS implementation performs equally or even better when I played the bagfile at 2.0 rate.
Can anybody spot some light on this behavior. ?
Any information regarding the restriction on the number of threads that can be used in ROS process.?