The Optimization Of The Distributed AI Embedded Cluster System

This paper improves the core functions of the real-time video analysis system for edge computing which we proposed before. The improvement of this paper including distributed cluster status management and two-level horizontal expansion. The three-level task scheduling mechanism is realized, and the complete data link processing is realized. At the same time, RK3399pro is proposed to replace RK3399 as the hardware computing unit of the AI business. On the premise of satisfying real-time processing, the computing unit resources are saved. Compared with our previous system, the processing capacity of the new system is improved by 6-8 times.