Lowering the impact of chunk losses in P2P live streaming systems

Abstract

Live streaming systems are becoming increasingly more popular. These systems attract a growing number of users, and some important TV channels already broadcast their live content on the Internet. Peer-to-Peer live streaming systems may provide a low cost solution to the problem of transmitting live content to a large number ofviewers. However, these systems are not fully reliable and do not implement quality assurance mechanisms, which leads to a degradation of system performance experienced by peers. This thesis studies chunk loss minimization in P2P live streaming systems, understanding the reason behind losses, and proposing new ways to respondto each reason while keeping low costs. We have developed our own fully capable P2P live streaming system and have used it to characterize chunk loss. We found a way to minimize chunk losses in freeriding scenarios by avoiding to request chunks to uncooperative peers. We proposed a support mechanism to recover "about to be lost" chunksthat should significantly reduce losses; it forwards emergency requests to sources outside of a peer neighborhood. We mapped this mechanism to a CDN-P2P system and developed an algorithm to dynamically minimize the number of video streams seeded to the P2P overlay while guaranteeing efficient dissemination and avoiding emergencyrequests. We have evaluated it under diverse conditions in which it may save up to 30% of upload bandwidth previously used by emergency requests without reducing stream quality.