Congestion Management: Using Deep transfer learning for Traffic Classification, Layer4 forwarding, OpenFlow meter

Abstract

Network traffic classification methods are used for congestion management, QoS delivery, billing in ISPs, and security purposes in firewalls. In the past, port-based, data packet inspection, and traditional machine learning techniques have all been widely utilized. However, their accuracy has decreased as a result of the Internet traffic's tremendous changes. The training data and test data are assumed to have independent, identical distributions by deep learning models used for network traffic classification. Due to changes in traffic features, this assumption could be incorrect in actual traffic classification. The classification of new network traffic will fail to use the models that were trained on the existing data. In this research, a deep transfer learning model without the aforementioned assumption is provided. The deep transfer learning technique is used to transfer the knowledge learned by the pre-trained traffic classification model to another model that has a smaller dataset and computational resources. The transfer learning model is built on ConvlD and BiGRU hybrid models that can achieve 98% accuracy in a completely new traffic classification target domain. In this research, deep transfer learning techniques are used to get over resource constraints and construct models for classifying network traffic based on deep learning. After classifying, the packets are forwarded by the OpenFlow switches according to the flow table configuration. Also, packets are forwarded by using Layer4 forwarding and OpenFlow metering methods for optimal bandwidth allocation to avoid network traffic congestion.