An Edge-Cloud Collaborative Computing System for Real-Time Internet-of-Things Applications

Xing Liu, Yongqi Yu, Mei Yu, Denghong Liao, Yaxin Li, Zihan Ji

Abstract: Edge-cloud collaborative computing system (ECCS) can combine the advantages of edge computing system’s low computing latency and cloud computing system’s high computing performance, which makes it widely applied in the real-time Internet of Things (IoT) applications. This paper presents an ECCS based on the open-source EdgeX and Huawei openLooKeng. To enable the ECCS to be low-latency, low-power and intelligent, the ECCS is integrated with a serial of enabling technologies such as lightweight k8s (k3s), heterogenous computing acceleration, edge intelligence, edge-cloud joint inference and federated learning. First, the EdgeX is built on k3s to make the ECCS have the functionalities of automating deployment, scaling, and management of containerized applications. Then, a set of algorithms such as AlexNet, YOLO and fast Fourier transform (FFT) are integrated into EdgeX to enhance the edge nodes’ intelligence and functionalities. Following that, several FPGA and GPU accelerators are developed and deployed on the edge side to accelerate the computationally-intensive tasks which run on the edge nodes. Finally, the joint inference and federated learning mechanisms are implemented to improve the algorithm accuracy as well as protecting the data privacy. The proposed ECCS has been implemented on the Zynq SoC and Raspberry Pi boards, and the real-world experimental results show that this ECCS has low resource cost, intelligent data processing capability, and also high real-time response performance.