Research On Proxy Caching For Multimedia Streams

Streaming media has become the primary service of today's Internet traffic. As an interdisciplinary study between multimedia and communication, the application and research have obtained rapid development. Streaming media application, international standard and foundation research are becoming the focus of industrial field and research institution. Streaming media researches consist of video codec, streaming server, peer to peer streaming and streaming media system technology. The typical streaming media system technologies include delivery policy, caching strategy and multicast mechanism. Not only integrates the three technologies above, but it also considers the performance optimization of the system. Therefore, it has been the focus and tendency of the stream media research.Based on the research of the sate-of-the-art solutions, this thesis proposes a dynamic shared buffers algorithm, a transmit policy which combines prefetching and patching, a closet reachable stream merging algorithm, a popularity-based segmentation method and a error resilience technology for MPEG-4 video transmission based on user data domain.Firstly, we survey the correlative technologies of the proxy-assisted media streaming from multimedia codec and transmission protocol to multiple proxies cooperation, introduce the existing problems and algorithms, classify, compare the primary solutions. All above the discussion and analysis is the basis of further research.This paper proposes a error resilience technology for MPEG-4 video transmission based on user data domain. It includes adaptive forward error correction, error feedback and error propagation prevention algorithms through user data domain. Simulation results show that the recovered video quality is obviously improved using this technique over variable channels. And the method is compatible with MPEG-4 bitstream syntax.Caching media objects in the memory of a proxy server helps to reduce network traffic, disk I/O bandwidth requirement and data delivery latency. There are some limits in the fixed-sized caching, interval caching, shared running buffers caching techniques. We propose a dynamic shared buffers algorithm (DSB) in the proxy to addresses these limits. The algorithm can adaptively allocates memory buffers and fully utilizes the caching data in streaming sessions, aims to reduce the memory scheduling and the network traffic.Segment caching generalizes the prefix caching algorithm by partitioning a media object into a series of segments, making caching decision by their respective utilities. Prefetching can avoids the jitter and patching can reduces network traffic. In the thesis, we integrate proxy caching with traditional reactive transmission schemes to develop a set of proxy-assisted delivery schemes and an adaptive segmentation approach that have achieved significant performance improvement and reduced significantly the consumption of aggregate bandwidth.Multicasting allows a media server to accommodate concurrent client requests with shared channels through patching or stream merging. In VoD system, Stream merging schedules lead to significant server and network bandwidth savings. Early stream merging algorithm reduces the cost of making an optimal merge tree, optimal stream merging is only can be used in offline system. This thesis proposes a closet reachable stream merging policy which achieves performance close to optimal offline hierarchical merging and better then ERM.The segmentation-based caching methods have greatly improved media caching performance. However, client accesses to media objects typically represent a skewed pattern. Most accesses are for a few popular objects and these objects are likely to be viewed in entirety or near entirety. Caching strategy with a predefined segment size, uniform or exponential, always favorably caches the beginning segments of media objects and does not resolve the fact that most accesses targeting to a few popular objects. The user access patterns are dynamically changing. The popularity of media object and most-watched portions may vary with time. In this situation, using a fixed strategy of caching several early segments my not work because this may overload the network during the initial time period as suffix segments need to be retrieved frequently; then during the later time, caching all the initial portions may become wasteful of system resources. The uniform or the exponential segmentation strategies use the fixed base segment size to partition all the objects. Without an adaptive scheme, an overestimate of the base segment size may cause an inefficient use of cache space while an underestimate may cause increased server management overhead. We propose a popular segment-based caching mechanism which considers the fact that most accesses are targeted toward a few popular objects or partials of an object. The popular segments are cached on the proxy and the size can be dynamic changed according to the client access behaviors in real time. The admission and eviction of segments are carried out using an accurate utility function better then the adaptive and lazy segmentation caching, improve client-perceived startup latency.