Abstract: The online games develop rapidly around the world and its number of users is still increasing. To analyze the characteristics of game traffic is of great importance. Wireshark is used to collect game traffic of different types, including Massive Multiplayer Online Role Playing Game（MMORPG）, First Person Shooting（FPS）, Real-Time Strategy（RTS）and cards, in order to study the recognition of game traffic. Then it filters the data with transport protocol and IP subnet and analyzes a number of statistical features for both upstream and downstream game traffic. Some features are effective for recognizing different game applications, including PPS, entropy of packet size and proportion of inbound to outbound data packets. Experimental results show that higher recognition accuracy can be obtained with the proposed feature combination and IP filtering.

Key words: classification of game traffic, statistical features, Internet Protocol（IP） filtering, Support Vector Machine（SVM）

摘要： 网络游戏在全球范围内迅速发展，其用户不断增加，因此游戏数据流的识别有着重要的现实意义。利用Wireshark，抓取不同类型的网络游戏流数据，包括大型多人在线角色扮演类、第一人称射击类、实时策略类游戏和卡牌类等游戏，对游戏流识别进行研究。采用协议过滤和IP过滤的方法对数据进行预处理，分别对下行和上行数据进行大量的统计特征分析，发现包大小信息熵、下上行包数目之比和PPS（Packets Per Second）等特征适用于游戏流分类。分类实验结果表明，利用IP过滤和提取出的特征组合可以有效地提高识别准确率。

关键词: 游戏流识别, 统计特征, 互联网协议（IP）过滤, 支持向量机（SVM）