Fast Dissemination for CommentCast in Unstructured Peer-to-Peer Networks

Abstract

Commenting is an important fundamental functionality offered with video stream- ing service. Users exchange commentaries and report problems about the video content through commenting. Tribler is, however, still lacking of the functionality of commenting. CommentCast is a fully distributed commenting system in Tribler based on the BuddyCast protocol stack. In this thesis, we improve the original design of CommentCast by augmenting different protocols working cooperatively to supply a fast, bandwidth-efficient and reliable commenting service. Then, we elaborate the protocol focusing on fast dis- semination. By comparing with gossip algorithm and flooding algorithm, we de- cide to use LightFlood [14], an algorithm combining pure flooding and spanning- tree broadcasting, to disseminate comments. Our experiments show that Light- Flood is a very fast and cost-effective dissemination algorithm. In order to study the knowledge of user commenting and evaluate our design, we collected the comment history of movie section of Verycd.com, a website providing P2P download resources. Since the movie section of Verycd.com has a similar context as Tribler, we take the collected comment history as a real workload for simulation of CommentCast. Finally, we investigate the performance of the push protocol of CommentCast by simulating the algorithm working under the real-world workload. Our simulation shows that the CommentCast is able to spread comments rapidly to a large number of users. At the same time, the bandwidth consumption is also realistic based on today’s network infrastructure.