I am using the pointcloudtolaserscan node, with concurrency level 0. The node is running in the Pepper robot, with an Atom E3845, quad core processor at 1.91 GHz.

The node is using 100% of the cpu and cannot keep up with the input point cloud at 10 Hz, with the output being published at less than 1 Hz.

If I use top to follow the process and the multiple threads, I see 16 entries for the pointcloudtolaserscan node. But only the first two use around 35% of the cpu, with the 3rd and 4th entries using around 10%, and almost no processing for all the other threads.

If I changed the concurrency level to 1, there are multiple threads, but only one doing all the processing, at almost 100% cpu usage. The output rate in this mode is almost the same.

In the end the multi-thread mode is using more or less the same amount of CPU time as 1 thread only. Why doesn't it use 100% of the cpu in each thread in the multi-threading mode?