| Title |
Method for Performance Isolation of Safety-Critical Applicationsin Mixed-Criticality Systems |
| DOI |
https://doi.org/10.5573/ieie.2019.56.12.27 |
| Keywords |
Deep neural network ; Object detection ; Mixed-critical system ; Performance isolation ; |
| Abstract |
Autonomous vehicles are representative mixed-criticality systems. In such systems, safety critical applications which are closely related to safe operation and must be guaranteed in real time, are performed by sharing system resources such as CPU or memory with non-safety-critical ones. Object Detection is the most important safety-critical application and shows high accuracy using DNN technology. However, due to the large amount of computation and memory usage, performance interference is inevitable when sharing system resources with other applications. In this paper, we propose a method to reduce system resource utilization of non-safety-critical applications when safety-critical ones are get started. The proposed scheme is running on top of NVIDIA Jetson AGX Xavier and its effectiveness is verified. Experimental results show that YOLOv3-tiny improves by up to 39% when the proposed technique is applied, even under being interfered by heavy memory-intensive applications. |