Live streaming in a hybrid CDN-P2P architecture

Abstract

Live streaming has the capability to generate large amounts of traffic since it is replacing traditional broadcast viewing. It is expected to become the fastest growing area of internet video due to the emergence of live video streaming offerings on multiple platforms such as YouTube, Twitter and Facebook. Most content providers have opted for direct distributions using Content Delivery Networks (CDNs) even though there has been a strong case for low-cost delivery using peer-to-peer (P2P) for many years. Reducing the traffic introduced from CDNs without compromising the service to the end-user has the capability of decreasing the expense to build new CDNs and allows CDNs to serve additional clients. One of the approaches that can accomplish the reduction in traffic is the utilisation of clients as supplementary sources of video when media content is delivered to them. A client that has already acquired the content can forward the media to its neighbouring peers, thus relieving the need for the CDN to serve these clients. This establishes a P2P network with clients downloading the content from other clients. Economical and extensive streaming can be provided using a P2P approach while enhanced performance streaming at the cost of higher prices can be provided using a CDN approach. A hybrid CDN-P2P architecture is a promising approach for achieving a compromise between streaming performance and system scalability. This work provides a hybrid CDN-P2P architecture which reduces work load on the CDNs, achieves low end-to-end delay (EED) and provides adaptive streaming. The work load on CDNs is reduced by utilising a hybrid tree-mesh P2P overlay. Low EED is achieved by considering key parameters of clients such as their bandwidth capabilities, geographical location and age when constructing the P2P overlay and by utilising CDNs to bring content closer to the users. Adaptive streaming is provided by using MPEG-DASH. HTTP/2 is used to utilise the server push functionality to further reduce the delay. The CDN-P2P model was compared with the traditional client-server model to present the savings which can be achieved by using P2P. The CDN-P2P model consumes 4.32 GB less average total CDN bandwidth than the traditional client-server model in 200 seconds of simulation time when considering 100 clients. Adaptive streaming achieves 6.95 seconds lower average EED when compared to single bitrate streaming and HTTP/2 attains 0.23 seconds of lower average EED when compared with HTTP/1 in 200 seconds of simulation time. The hybrid P2P overlay presented in this work was compared to a mesh overlay and this work obtains 4.2 seconds lower average EED once 100 clients join the overlay.