基于爬行器的大规模P2P IPTV测量

摘要

为了解大规模P2P IPTV系统中的用户行为特征和拓扑结构特征等内在信息,开发和部署了一个多协议P2P IPTV爬行器TVCrawer,对3个主流的P2P IPTV系统--PPLive,PPStream和UUSee进行了大量的主动测量,并对P2P IPTV系统中的用户行为和网络拓扑特征进行了分析和比较.主要发现包括:1)P2P IPTV系统的频道在线人数中,有一半以上位于不可达的NAT或者防火墙后面;2)节点动态性的波动范围随频道人数的增加而增加,但是其取值范围具有幂律上限;3)节点会话长度符合广延指数分布;4)PPLive的入度分布属于具有指数截断的幂次分布,PPStream的入度表现为某种分段幂律函数,UUSee的入度接近威布尔分布;5)P2P IPTV系统都是异配网络;6)P2P IPTV系统都表现为小世界网络;7)PPLive网络具有聚类特征,而PPStream和UUSee则不存在明显聚类特征;8)3个系统都表现出类似于无标度网络的鲁棒性特征,而与其他两个系统比较时,PPLive具有更高的故障容错性和更明显的攻击脆弱性.这些测量研究和发现不仅有助于设计出更符合真实网络应用环境的系统或协议,也是实现对P2P IPTV进行监测,引导,控制等方面的重要依据和基础.%In order to gain insight on the topology characteristics of P2P IPTV systems and their online user behavior characteristics, This paper develop and deploys a multiprotocol P2P IPTV network crawler, called TVCrawler, which enables users to launch an in-depth measurement and comparative research for several well-known P2P IPTV systems, including PPLive, PPStream, and UUSee. This paper presents results from experiments and research efforts on these large-scale P2P IPTV overlay graphs. Major findings include 1) more than 50％ of online users are unreachable because they lie behind NAT or firewall; 2) fluctuation range of churn increases with thc increase of the population of channel, and there exists a power-law upper bound for fluctuation range of churn; 3) session length of peer follows a stretched exponential distribution; 4) while the in-degree of PPLive follows the power-law distribution with exponential cutoffs, the in-degree of PPStream appears to have multiple separate power-law regimes with different exponents, and the in-degree of UUScc approaches Weibull distribution. 5) All of measured P2P IPTV overlay graphs appear as a disassortative mixing; 6) small-world networks are ubiquitous in measured P2P IPTV systems. 7) the PPLive overlay graph is clustering, but the others are not; 8) all of measured P2P IPTV overlay graphs have robustness similar to that of a power-law graph, but PPLive is more vulnerable to malicious attacks and more robust to random failures than the other. These research and findings Will be good for designing and optimizing of P2P IPTV protocols, but also for monitoring, directing,and dominating the P2P IPTV system.