Network Caching for Efficient Multimedia Content Delivery

指数增长的多媒体信息内容已经成为现代通信网络的核心问题。这是传统通信网络的一个范式转变，在传统通信网络中，主要功能是信息传输，通过更好地确定带宽和交换资源，可以实现服务质量的改善。然而，在信息存储起着核心作用的网络中，如果没有信息的缓存、预取和/或镜像的帮助，服务质量的重大改进就不可能成功。可以说，即使有无限的免费带宽可用，信息集中在少量服务器上也会导致服务器过载，导致不可接受的下载延迟。 这个普遍的问题已经创造了爆炸的研究，是要在网络系统工程文献中发现的Web缓存。虽然这项工作已经贡献了重要的工程解决方案，但其中大部分是临时和非正式的。事实上，在大量关于缓存研究的文献中，很少能找到涉及随机基础的论文，特别是那些支持随机模型的论文。此外，很少有人发现证明在严格设置的非平凡的随机或平均情况下的缓存结构和算法的属性。我们的论点是，需要采取更系统的办法。这个提议描述了一个项目，以满足这一需求;更一般地说，我们概述了一个系统的治疗，侧重于基本的设计问题，一个，在处理这些问题，集成实验，分析和统计测量。 网络缓存的设计目标是减少访问延迟、流量拥塞和服务器负载。这些目标只能通过一个深思熟虑的设计来实现，解决了许多重要的和具有挑战性的研究课题。一些基本的问题，仍然没有明确的答案包括：动态缓存和缓存与调度时间;设计和分析容易实现的启发式算法;影响本地请求序列的缓存性能;缓存和预取在低带宽访问环境中，特别强调无线Web访问;缓存分配和sizingproblems;和Web图形的性能建模。扩展知识库和加深我们对这些和其他几个同样基本的缓存系统设计问题的洞察力是本提案的主题。最终的目标是利用这一改进的知识库开发一个实验测试床，实现实际可行和高效的网络缓存系统。 这里确定的研究主题的内在复杂性，以及网络缓存问题作为一个整体，必然会调用所有可用的研究工具。因此，从方法论上讲，拟议的研究范围从数学建模和分析到统计测量和实验。我们的研究结果的影响，其interdisciplinary性质，将不仅限于设计多媒体网络缓存系统，但将可能导致改进的问题解决技术在相关领域的计算机算法，软件工程，概率论和运筹学。

Exponentially increasing multimedia information content has become a problem at theheart of modern communication networks. This is a paradigm shift from traditional com-munication networks, where the primary function is information transmission, and wherequality of service improvements are attainable through better dimensioning of bandwidthand switching resources. However, in networks where information storage plays a centralrole, major improvements in quality of service can not be successful without the help ofcaching, prefetching and/or mirroring of information. It can be argued that even if anunlimited amount of free bandwidth were available, the concentration of information on asmall number of servers would cause server overloads, resulting in unacceptable downloadlatencies. This general problem has created the explosion of research studies that are to be foundin the network systems engineering literature on Web caching. Although this work hascontributed important engineering solutions, much of it is ad hoc and informal. Indeed, inthe large literature on caching research, it is rare to find papers dealing with mathematicalfoundations, especially those underpinning stochastic models. Also, one rarely finds proofs ina rigorous setting of nontrivial stochastic or average-case properties of caching structures andalgorithms. It is our thesis here that more systematic approaches are needed. This proposaldescribes a project to meet this need; more generally, we outline a systematic treatmentthat focuses on fundamental design issues, one that, in dealing with these issues, integratesexperimentation, analysis and statistical measurements. Network cache design objectives are reductions in access latency, traffic congestion, andserver loads. These objectives can be attained only through a thoughtful design that ad-dresses many important and challenging research topics. Some of the fundamental questionsthat remain without definitive answers include: dynamic caching and caching with expirationtimes; design and analysis of easily implemented heuristic algorithms; impact of locality inrequest sequences on caching performance; caching and prefetching in low bandwidth accessenvironments, with a special emphasis on wireless Web accesses; cache allocation and sizingproblems; and Web-graph performance modeling. Extending the knowledge base and deep-ening our insight into these and several other equally fundamental caching system designproblems is the main theme of this proposal. The ultimate goal is to utilize this improvedknowledge base to develop an experimental testbed for achieving practically feasible andefficient network caching systems. The inherent complexity of the research topics identified here, and the network cachingproblem as a whole, necessarily call into play all available research tools. Thus, methodologi-cally, the scope of the proposed research ranges from mathematical modeling and analysis tostatistical measurements and experimentation. The impact of our results, by their interdis-ciplinary nature, will not just be limited to designing multimedia network caching systems,but will potentially lead to improved problem solving techniques in related fields of computeralgorithms, software engineering, probability theory, and operations research.