Research On MPEG-DASH Flow Rate Control And Cache Optimization

With the development of Internet streaming services,people raise higher requirements on the viewing experience of Internet audio-visual contents.MPEG-DASH,a dynamic adaptive streaming technology based on HTTP protocol,can effectively utilize the network bandwidth resources to provide high-quality viewing experience,along with good versatility and deployability.Such features lead to wide application of MPEG-DASH in mainstream Internet streaming services.Meanwhile,proxy caches and CDNs are widely used in multimedia distribution to improve users' experience and reduce the load of backbones.This technical trend imposes great impact on the development of IP networks.The adaptive bitrate(ABR)selection algorithm on the user's side plays core role of DASH protocol,affecting users' experience significantly.However,existing ABR selection algorithms do not take into account proxy caches which are located at the edge of access networks,resulting in susceptibility to proxy caches' interference.This susceptibility can further leads to the bitrate oscillation(Ping-pong Effect)and video stalling.In addition,the existing multimedia cache management strategies cannot effectively tap the characteristics of DASH streaming media,resulting in unsatisfactory performance.Therefore,it is of great significance to develop better proxy caches for DASH streaming media distribution.This thesis focuses on the issues of DASH streaming media distribution,and studies the relationship between the DASH protocol and the proxy caches.Afterwards,a cache-optimized ABR selection algorithm and a DASH proxy cache management strategy based on grey models are proposed.The main contributions are as follows:(1)By analyzing the effects of proxy caches,this thesis declares that the existence of proxy caches might lead to the fact that there are two sources for each DASH slices.Owing to this "dual-source" phenomenon,existing ABR selection algorithms,categorized as rate-based and buffer-based ones,are susceptible to proxy caches' interference,leading to bitrate oscillation and video stalling.To solve this problem,a cache-optimized DASH ABR selection algorithm called COBA is proposed,which makes use of overlapped buffer sections to eliminate the bitrate oscillation and video stalling.COBA adopts an additional fast start algorithm to accelerate the convergence of video bitrate.(2)Facing the challenge that the popularity of DASH slices is hard to predict,this thesis proposes a grey model based DASH proxy cache management strategy,GPCM,and designed the architecture of the proxy cache.GPCM consists of two parts:cache prefetching strategy and cache replacement strategy.The core of GPCM is the popularity prediction of DASH slices based on grey models,which features short-term prediction accuracy and less time-consumption.The cache prefetching strategy performs prefetching bitrate calculation and prefetching operation,while the cache replacement strategy selects the slices to be replaced.These two strategies both make decisions according to popularity prediction results of slices.(3)Extensive experiments were conducted to evaluate the performance of COBA algorithm and GPCM strategy.First,a real experiment scenario is built using FCC throughput traces to simulate real network throughput,and the performance of COBA algorithm was tested.Afterwards,to address the large-scale test requirement for GPCM strategy,a proxy cache simulator called DashProxySim was developed.The performance of GPCM strategy was evaluated via DashProxySim driven by trace datasets.Experimental results show that the COBA algorithm can eliminate the bitrate oscillation and reduce the video stalling of DASH clients.The GPCM strategy improves the cache hit rate and clients' throughput.By combing COBA and GPCM together,the users' viewing experience can be significantly improved.