SMILER: Towards Practical Online Traffic Classification

摘要: Network traffic classification is extremely important in numerous network functions today. However, most of the current approaches based on port number or payload detection are becoming increasingly impractical with appearance dynamic encrypted applications. Even though some supervised learning work were proposed, it difficult to collect sufficient flow-labeled traces for training. On other hand, online needs an early identification, which still challenging well-known approaches. In this paper, we propose a semi-supervised approach named SMILER, supports from sizes first few packets (empirically 5 packets) flow. Experiments real networks demonstrate that SMILER achieves 94% precision and 96% recall average all tested applications, even disordered works well. With hybrid scheme, performance further improved. Meanwhile, performs fast both updating. All experimental results show practical accurate classification.