Measurement Analysis And Modeling Studies Of P2P Media Streaming Applications

As the online video applications develop rapidly, the traditional C/S (Client/Server) structure emerges some problems of poor scalability, low reliability and server bandwidth bottleneck. The P2P media streaming system based on distributed content can effectively solve the above problems because it makes the most of peer bandwidth and storage resources. However, some problems in P2P media streaming systems still need to be solved. From the server side, the problems include how to further reduce server bandwidth consumption and how to ameliorate system protocol design; From the client side, the problems include how to support fluent playback of large number of high bit rate VBR(Variable Bit Rate) videos, how to reasonably plan system resources according to user behaviors and how to optimize system application in mobile environment. To deal with above problems, the paper does deep theoretic analysis through large scale measurement. The main works are as follows:(1) Existing P2P rate based downloading strategy (downloading state switch mechanism) has shortages of poor playback continuity and relatively high server bandwidth consumption which is subject to the impact of P2P rate fluctuation. To deal with the issue, we come up with a new state switch rule based on local buffer amount. We develop a model using Brownian Motion and gain the relation expression of the ratio of saved server bandwidth, playback bit rate and downloading rate. Theoretical analysis, numerical evaluation and measurement in real-world system show that our mechanism can effectively alleviate the negative impact of P2P rate fluctuation on the ratio of saved server bandwidth, as well as provide better playback continuity. Furthermore, it can be derived that the ratio of saved server bandwidth in the new mechanism is particularly related with the ratio of playback bit rate to the HTTP downloading rate. This implies important theoretic reference for further saving server bandwidth.(2) To deal with the issue that more and more high bit rate VBR videos are published in P2P VoD system, which degrades the comprehensive playback quality, this paper uses theoretical model to make out the relation expression of downloading rate, playback bit rate and playback fluency and points out that increasing the initial video prefecthing amount can improve the playback fluency. Furthermore, this paper compares the playback performance of VBR vs. CBR(Constants Bit Rate) videos and high vs. low bit-rate videos as well as their requirement for the sharing environment. This paper theoretically makes out the relation of various parameters of different kinds of videos and points out that in order to reach the same playback fluency with CBR videos, VBR videos need much more download bandwidth due to the dynamics of playback bit rate. This work may provide theoretical reference for system optimization.(3) Mastering the features of the current large scale P2P streaming systems is the basis to optimize this kind of systems. PPLive and PPStream are two most popular P2P streaming systems in China. The protocol of PPLive has been fully studied before, while that of PPStream lacks transparency. To deal with the issue, this paper figures out the buffer structure and sharing mechanism of PPStream system through measurement and data analysis. It is found that PPStream has jumping sharing window. Further, this paper uses "window jumping model" to analyze the impact of buffer structure and sharing mechanism on peers'reciprocal relation in PPStream system and also does deep comparison study with PPLive. It is found that different with PPLive which possesses sliding sharing window, the jumping property of sharing window in PPStream can effectively restrict the behavior of users prefeching video contents excessively which may rob other peers'bandwidth resources. So this property can effectively stabilize playback quality. These studies in the paper make good references to further optimize the system protocols and improve its playback quality.(4) In P2P streaming system, since users download as well as provide video contents, their behaviors have strong impact on system performance. Nevertheless, existing relative studies are mainly based on IPTV systems which use IP multicast technology, while rarely focus on P2P streaming system. To deal with the issue, this paper uses measurement data in realworld system to analyze the arrival and departure features of users in different types of channels and also the behaviors of users staying watching one program in a channel. It is found that (A) user behaviors exhibit different features in different kinds of channels, which are influenced by user daily routine and channel type;(B) users watch videos actively within2--3minutes after they arrive at one program, however once they stay over10minutes, they will leave uniformly. In addition, the behavior of watching programs successively is memoryless and influenced to a large extent by program arrangement. These analyses make sense for soundly planning system resource and reasonably arranging advertisement.(5) As the mobile internet develops rapidly, users desire for mobile P2P streaming application becomes more and more intense. However, existing P2P streaming systems are mainly based on fixed internet, thus it is unkown how the system can support peers mobility. To deal with the issue, this paper explores the influence of one mobile peer on the system parameters. This paper develops a virtual mobile environment through changing private IP address. Then the paper studies the influence of peer mobility on two P2P streaming systems in terms of service interruption time (mobile peer has no data interactivity with other peers during the time), neighbor topology variance and playback progress. The analyses are sure to bring good insight on the design and optimization of mobile P2P streaming system.