Toward Improving Scheduling Strategies in Pull-Based Live P2P Streaming Systems

Abstract

Several recent P2P streaming systems have adopted mesh overlays to disseminate content to participating peers because this topology appears to be more resilient to churns. To cope with inferred problems, such as data redundancy, these systems opt for data-driven content retrieval mechanisms (pull mechanisms). Each node has a list of neighbors with whom it periodically exchanges buffer information and requests content fragments. One of the drawbacks of such a mechanism is that it does not offer intelligent selection of sending neighbors based on their characteristics. This is mainly because the most important criteria used to select nodes is the content availability. This can result in some performance degradation, for instance, due to peers that are sending very small or big parts of the needed data. Resiliency may then be weakened and overhead increased. In this paper we propose to study how the integration of some end nodes characteristics can improve the performance of a typical pull mechanism with random scheduling. We show that the improvement in performance is significant enough despite the fact that the room for improvement is bounded by the limitations of the pull mechanism. Hence we believe that the awareness of end characteristics is an important block upon which we can build more efficient content retrieval mechanisms. The gain in performance can also be amplified by proposing an alternative to the pull mechanism such as a combined pull-push approach.