18.0.1
Description
Abstract
Stop using multiple thread for ray-tracing to improve performance
Background
In #886, ray-cast parallelization was implemented to improve performance.
However, currently, due to resource exhaustion caused by the increase in the number of Autoware nodes, it is better to not parallelize it.
Details
Perforcemance measurement
CPU1 : Intel i9 14900K
CPU2 : Intel i7 10510U
Autoware version: Versions used for regression testing(version Sep. 9th)
scenario: arounding 5 cars
The ray-tracing time mean time comparison before and after pull-request
| CPU 1 | CPU 2 | |
|---|---|---|
| Before | 590.1 μs | 10140.1 us |
| After | 416.4 μs | 4744.82 us |
The measured value is the average execution time of the point cloud generation processing part for 20 seconds from the start of the scenario, excluding the first 10 seconds.
References
RegressionTest: OK
Destructive Changes
None
Known Limitations
None