Towards intelligent P2P IPTV overlay management through classification of peers

IPTV has emerged as one of the popular Internet applications attracting immense interest of academia and industry. Among others, Peer-to-Peer (P2P) approach enables IPTV service with ease of deployment and low cost. P2P systems involve end-hosts, called peers, to share their resources for dissemination of stream. In these systems, user activities, such as join and quit operations, translate as activities of peers. Due to this reason, these systems become highly dynamic as the behavior of users has a significant impact on the stream delivery performance of the whole system. Earlier P2P approaches deal with user behavior indirectly through enabling resilience in the system, which leads to other issues such as increased latency. Latter approaches attempt to address user behavior directly through proposing user behavior models. Such models focus to learn and predict user behavior in order to adapt the overlay network accordingly. Towards this, one important aspect is the classification of users. However, machine-learning classifiers have not been extensively evaluated for their accuracy over the classification problem. In this paper, we extensively evaluate numerous classifiers for their accuracy over different parameters. These results may be used to design intelligent P2P overlays for IPTV services providing better performance in terms of efficient stream delivery to users. Our results show that the Decision tree classifier performs better than other classifiers for this particular problem.