Peer assisted multicast streaming on-demand applications

Abstract

On-demand multimedia streaming applications allow users to select and view content at a time of their choosing and to control the playback of this content. This level of control typically requires that each user receives a dedicated stream from the service provider. The provision of these streams can place a high load on a centralized server and can result in network bottlenecks near the server. An approach called Peer Assisted Multicast Streaming is proposed to reduce the load on a centralized server, and to partially re-distribute the remaining load to the edges of the network. Peer Assisted Multicast achieves this by allowing participating users to share with each other the beginning, or “prefix" of content items that they have previously received and cached, thus creating a decentralized collaborative cache. By doing this, a server will only need to serve the remainder of the content and can delay doing so until the content is required. This content can be provided using a multicast stream. Other requests received before the multicast stream begins can be serviced by the same stream. This allows the server to benefit from the use of multicast without affecting an individual user's experience. Serving multiple requests with a single stream will reduce the overall network bandwidth requirements of a server. As the users of the system will be providing content to each other, some of the network load will also be distributed away from the server. Peer Assisted Multicast Streaming can be implemented using standard network protocols and existing peer-to-peer architectures. An implementation using the RTSP stream control protocol and the Chord overlay network is proposed. Using a detailed network simulation Peer Assisted Multicast Streaming was evaluated under varying conditions. In the evaluation, Peer Assisted Multicast Streaming was shown to provide a 21.62% reduction in the number of concurrent multimedia streams required by the server when a prefix of 16.67% of the duration of the content was provided by the users. It was established that for the most popular content the maximum number of concurrent streams was independent of the request rate of the system, which is not the case for unicast based approaches. Peer Assisted Multicast Streaming was also shown to respond well to the sudden introduction of new popular content.