Master Thesis - 2010

Networks with sufficient bandwidths provide various content distribution services. In particular, a large number of users demand to distribute their contents. Peer-to-peer is one of promising technologies to improve scalability. As existing content distribution systems focus on business services, transmission load of distributor tends to be heavy to keep quality of service higher. P2P live-streaming distribution has been proposed to reduce load of distributors. There are various types of contents with appearance of live-streaming distribution services. It is difficult for users to select their desirable contents immediately. Thus, the system assisting user's content selection is necessary. In traditional streaming systems, special terminals need to collect information for content recommendation in order to assist interesting content selection from multiple contents. Moreover, in live-streaming distribution systems, distributing contents change frequently. Meta-information or thumbnails does not suitable for recommendation technique in P2P live-streaming system.

Under these backgrounds, this thesis proposes P2P live-streaming system considering transmission bandwidth of each node to utilize network bandwidth efficiently. This thesis also proposes new P2P live-streaming system selecting and distributing recommended contents on time based on users' similarity with access history. The system recommends and distributes low-definition video of content. The system can improve users' satisfaction with content recommendation and simplify play-back process of content.

The evaluation of the proposed system presents that the system reduces distributor's load drastically to distribute contents stably. The evaluation shows churn resilience of nodes in the system. The recommendation system also improve user satisfaction with low load of recommended distribution.