Abstract-Nowadays, Peer-to-Peer (P2P) live streaming applications have attracted great interest. Despite the fact that numerous systems have been proposed in the past, there are still problems concerning delay and quality requirements of live video distribution. In this paper, we consider a pull-based P2P live video streaming system where the video is disseminated over an overlay network. We propose a mesh-based overlay construction mechanism that enhances the received video quality while minimizing, as much as possible, the play-out delay. The principal feature provides each newcomer with a set of neighbors holding almost the same video segments and enough available transmission capacities to deal with its requests. A particular algorithm has been designed to estimate the peer's available capacities. The results of simulations show our mechanism efficiency in heterogeneous systems.Keywords-Peer to peer networks; live video streaming; overlay network; play-out delay.

Multimedia applications will need high speed networks. When designing dedicated servers for these new applicatwns, it is necessary to estimate their performance. The work presented here is a common work between CCETT (Centre Commun #Etudes de Tdlkdiffwwn et de Tkldcommunications) and INT (Institut National des Tdlkcommunications). It is a performance evaluation of a multimedia server, a device able to offer audiovisual sequences to several users simultaneously, through a high speed network (such as an ATM network). A queueing network model has been developed and results are analyzed. Many simulations allowed the optimization of the dimensioning and the implementation of the server, depending on several parameters. A prototype has been built. Presentation of the server IntroductionMultimedia information systems are only beginning to appear, but are still limited to local area networks or lack either visual or audio quality [2]. Totally digital, integrated, long-distance multimedia services are under development. Industrials and governments are fiercely fighting to impose their standards, because of the huge commercial interests at stake. Such applications as video electronic mail or audiovisual sales catalogues or even interactive television are expected. The first step is to obtain a video sequence of good quality, i.e. in color and with 25 or 30 frames per second. Significant progress will have been achieved when it is possible to transmit this kind of digitized video sequence through non specialized networks towards non specialized workstations.Research is already being conducted in order to specify how objects that are more complicated than video sequences (hypermedia objects) could be held [3].The data transfer needs for these multimedia services are very great. If high speed networks seem able to meet this need, the present technology of storage means (above all, hard drives) is too limited. Even with unexpected technical progress, a single hard drive will not deliver a sufficient throughput, not to mention the necessary storage capacity. As a consequence, the concept of parallelism will forcibly be used in designing future videocommunication appliances [4,5,6]. Parallelism allows several modules to work at the same time and to cooperate, thereby increasing the global power of high technology systems. However, as a result of distributing the worklaad over many components, new constraints are introduced : synchronization, concurrency, ... We are planning to develop a multimedia server with a very high data rate, using parallel architecture. It must be able to offer audiovisual sequences (videocommunication is CCElT's main vocation) and it will exploit the large bandwidth of high speed networks. The quantitative analyses we derived from our performance evaluation have allowed us to build a prototype of the server. Based on our evaluation and on prototype tests, the model will be extended so as to meet the requirements of a real commercial product.This study is a conceptual modelling based on the behavior of th...

No abstract