With the rapid growth of the demand on video streaming services, Internet has witnessed the great success of the Peer-to-Peer (P2P) live streaming systems. In the past several years, P2P live streaming architectures have advanced in two major approaches: pull-based approach versus tree-based (push-based) approach. Since they both have inherent performance limitations, recently, researchers are exploring a new class of streaming architecture, called the pull-push hybrid approach. In this work, we thoroughly study this new type of P2P live streaming architecture including models, algorithms and implementations....[ Read more ]

With the rapid growth of the demand on video streaming services, Internet has witnessed the great success of the Peer-to-Peer (P2P) live streaming systems. In the past several years, P2P live streaming architectures have advanced in two major approaches: pull-based approach versus tree-based (push-based) approach. Since they both have inherent performance limitations, recently, researchers are exploring a new class of streaming architecture, called the pull-push hybrid approach. In this work, we thoroughly study this new type of P2P live streaming architecture including models, algorithms and implementations.

The sub-stream scheduling problem is the key design issue in a P2P hybrid live streaming system. In this work, we propose a max-flow model for unifying this sub-stream scheduling problem. We find that the sub-stream scheduling problem in CoolStreaming+, GridMedia and LStreaming can be formulated into a special case of the proposed max-flow model. We demonstrate that the sub-stream scheduling module in these three hybrid systems actually solves one special case of the proposed max-flow model, respectively. Our proposed max-flow model can be solved in polynomial time and provides useful insights for a better sub-stream scheduling design, in which we propose a weighted max-flow scheduling scheme to better utilize heterogeneous peers.

In large-scale P2P networks, since participating peers reside behind different ISPs and use various methods to access the network, the networks always have apparent peer heterogeneity issue in terms of the upload bandwidth. Recently, the layered coding has become a viable and promising way to deliver real-time streaming traffic. However, existing layered streaming systems all adopt the pull-based streaming architecture, and the strength of the layered coding is not brought into full play in these systems. In the thesis, we point out the benefits of combining the hybrid architecture and the layered coding together, and demonstrate how to conduct the sub-stream scheduling in the layered streaming case, with the consideration of the layer dependency requirement.

In dynamic P2P networks, how to deploy a robust live streaming system with good streaming performance is the ultimate goal of the system design. In this work, we introduce the design details of our P2P hybrid live streaming system, LStreaming. LStreaming is controlled in a closed-loop feedback fashion to achieve automatic load balancing with rapid convergence so that it is robust in dynamic P2P network environments. Further, we introduce the enhanced version of LStreaming, which is called LStreaming+. We implement our weighted max-flow scheduling scheme in LStreaming+. Through our real network experiments, we find that LStreaming and LStreaming+ can achieve excellent streaming performance in terms of throughput, watching quality and overhead.