A green computing method for encrypted IoT traffic recognition based on traffic fingerprint graphs

The Internet of Things (IoT) is a revolutionary technological trend that is reshaping the way people interact with the digital world, creating limitless possibilities for future life and work. With the exponential increase in the number of IoT-connected devices, the substantial traffic data generated by smart devices poses significant challenges to traffic recognition, especially with the emergence of encrypted traffic for privacy protection. Encrypted traffic recognition is a prerequisite for ensuring Quality of Service (QoS) and network security. However, existing methods suffer from inadequate accuracy in encrypted traffic recognition and fail to address the issue of green computing of model training in IoT scenarios. In this paper, we propose a fast encrypted traffic recognition method, TFPGM, which is based on traffic fingerprint graphs to address these challenges. The core idea of TFPGM is to utilize interactive modes to represent the fingerprints of similar traffic types. TFPGM consists of two main phases: constructing traffic fingerprint graphs and establishing recognition models. In the construction phase, we analyze the interactive modes of encrypted traffic and represent them using undirected graphs. In the model-building phase, a dedicated recognizer is established for each traffic fingerprint to enhance the exploration of interactive modes within the same traffic class. These recognizers are combined to create a complete traffic fingerprint recognition model. In both phases, considerations were given to reducing the size of the graph and lowering the model’s complexity with the aim of achieving green computing in real-world scenarios. The results indicate that TFPG can complete model training in just 3.83 seconds while ensuring a 97.72