流媒体系统中变动分段缓存策略的研究

发布时间：2018-02-28 16:06

本文关键词： 流媒体分段方式超几何分布静态调度缓存替换策略　出处：《北京交通大学》2015年硕士论文　论文类型：学位论文

【摘要】：流媒体系统在实际中应用广泛,但大量图形数据在网络传输时出现响应速度慢、网络拥塞严重、缓存利用率低等问题。尤其在硬件带宽较低情况下,如服务器网卡读写能力不足、PCI时钟频率低、总线位宽数低于64位等情况下。本文针对该类问题提出一种基于预期使用价值的缓存替换策略,并以基于斐波那契数列划分的变动分段方式为基础实现流转化。本论文主要分为三个部分：第一,从流媒体关键技术出发,通过对其在不同传输协议、不同编码方式下数据段平均接收分段数、平均下载时间进行统计,分析调度策略在流媒体数据传输中的重要性。通过评估C/S网络结构得出其作为缓存替换策略测试平台的合理性。此外,本文也阐述了静态调度算法的分段方式在调度策略中和缓存替换策略成果及其在点播系统中的重要意义。第二,本文为预期使用价值算法的实现提出一种将斐波那契数列相结合的数据分段方式,并通过Ns2仿真软件对新的数据分段方式进行模拟统计。分析中发现新算法在减小对磁盘资源的占用和降低延时方面都比其余算法具有更好的效果。然后从用户点播体验和用户行为特征入手提出基于预期使用价值的缓存替换策略。第三,针对基于预期使用价值算法,通过二次编码Ns2仿真环境的底层C++代码来拓展Ns2仿真环境功能模块,对编译分级结构中TclObject基类添加派生类。通过建立哑铃型网络拓扑结构,分别对服务器响应延时、客户端请求延时、目标节点延时、网络吞吐量等指标进行测试,得出ULH算法优于传统的LRU、LFU算法。最后将全部策略嵌入到实际网络平台中,对流媒体播放效果进行实验,得到了较好的图像传输效果。
[Abstract]:Streaming media system is widely used in practice, but a large number of graphics data in the network transmission problems such as slow response speed, serious network congestion, low cache utilization, especially in the case of low hardware bandwidth. For example, the server network card is not able to read and write, the PCI clock frequency is low, the bus bit width is less than 64 bits, and so on. This paper proposes a cache replacement strategy based on the expected use value for this kind of problems. And based on the Fibonacci series division of the variable segmentation as the basis for the realization of flow transformation. This thesis is divided into three parts:. First, according to the key technology of streaming media, the average number of data segments received and the average download time under different transmission protocols and different encoding methods are counted. The importance of scheduling strategy in streaming media data transmission is analyzed, and the rationality of C- / S network structure as a test platform for cache replacement policy is obtained by evaluating the structure of C- / S network. This paper also describes the achievements of static scheduling algorithms in scheduling policies and cache replacement policies and its significance in on-demand systems. Secondly, this paper proposes a data segmentation method which combines Fibonacci series for the realization of the expected use value algorithm. Through the Ns2 simulation software, the new data segmentation method is simulated and counted. It is found that the new algorithm is more effective than the other algorithms in reducing the cost of disk resources and the delay. Based on the characteristics of broadcast experience and user behavior, a cache replacement strategy based on expected use value is proposed. Thirdly, based on the expected use value algorithm, the function module of the Ns2 simulation environment is extended by the lower C code of the secondary coding Ns2 simulation environment. By establishing the dumbbell network topology structure, we test the server response delay, the client request delay, the target node delay, the network throughput and so on. It is concluded that the ULH algorithm is superior to the traditional LRUL-LFU algorithm. Finally, all the strategies are embedded into the actual network platform, and the streaming media playback effect is tested, and the better image transmission effect is obtained.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TP333

【参考文献】