Scholarly Article
Real-time Processing Optimization of Convolutional Neural Network in Edge Computing System
Lai, Shean
2024-12-16 · Journal of Computational Systems and Applications · Cultech Publishing Sdn. Bhd.
Abstract
On the edge nodes of Azure Stack Edge, network interfaces often experience packet queuing during peak periods due to congestion and high load. To address this issue, MobileNet was used to analyze network traffic in real-time and detect congestion, aiming to balance node network load and reduce latency. Firstly, data was collected and key features were filtered during peak periods using Wireshark, including packet rate, bidirectional packet count, flag statistics, etc.; next, TensorFlow Lite was used to prune and lightweight the model, removing redundant network layers and reducing floating-point operation accuracy, making it suitable for edge devices; then, based on the optimized model, real-time classification of network traffic was performed using Apache Kafka to detect potential congestion situations; at the same time, the load balancing strategy was dynamically adjusted based on the detection results to optimize queue priority and traffic transfer, thereby alleviating congestion. The results showed that during peak hours, the average processing time for ICMP (Internet Control Message Protocol), TCP (Transmission Control Protocol), and UDP (User Datagram Protocol) traffic was reduced by 27.1%, 23.1%, and 20.7%, respectively. When the cache capacity was 16GB, the hit rate reached 94%. Under standard traffic conditions during peak hours, the packet loss rate dropped to 2.1%, which could significantly improve the performance of edge node networks.
Keywords
Edge computing, Convolutional neural networks, Real-time processing, Network traffic analysis, Apache Kafka
Citation Details
Journal of Computational Systems and Applications, pp. 1-14