Feature Analysis And Recognition Of Network Game Traffics

Recently, the popularization of online game is getting higher and higher. Classification of game traffic is of great importance to provide a stable network services and efficient networker source management. Moreover, it can also be used in the area of Quality of Service, service differentiation and billing. Choosing effective features is essential to the classification of game traffic based on the statistical features.In this thesis, the author selects HearthStone, XYQ, AssaultFire, DOTA, LOL and DOTA2 as research subjects. We perform an analysis of 7 features, including ratio of inbound to outbound data bytes, ratio of inbound to outbound data packets, packet size, BPS, PPS and IP subnet. A method for data filtering based on IP subnets is proposed to reduce the interference of non-game data, making the distribution of featurea more aggregated. According to the detailed analysis of the distribution of features, combination of PPS and entropy of packet size is founded to be able to classify HearthStone, XYQ, LOL and DOTA2 efficiently and accurately, while AssaultFire and DOTA share the same area which is different from the others. For AssaultFire and DOTA, ratio of inbound to outbound data packets distributes in different areas. Therefore, we use PPS, entropy of packet size and ratio of inbound to outbound data packets to classify and identify the game traffic. At last, SVM algorithm is used to perform experiments. The results show that using only 3 features and relatively small number of training samples can achieve higher classification accuracy with the proposed algorithm.