Diagnosis and Control of Network Variability by Massively Accessed Servers

这项提案寻求资金，以在网络感知服务器协议和服务的一般框架内调查一些基础研究问题。解决这些问题对于互联网的持续可扩展性至关重要。海量服务器是流行的互联网服务器，它们产生了流经网络的流量的很大一部分。目前，这样的海量服务器被调优以优化其自身的性能(例如，容量、延迟、I/O利用率)，而忽略了监控网络状况和有利地使用该信息的机会。这些信息既可用于改善网络行为(例如，缓解拥塞导致的问题)，也可用于进一步优化服务请求的性能。因此，海量服务器具有独特的定位(1)通过跟踪它们生成的流来观察和诊断网络状况，以及(2)通过更好地调节和调度它们注入到网络中的流量来管理和控制网络资源。必须在广泛的时间范围内追求这些目标，以最大限度地发挥可实现的有益影响。在最短的时间尺度上，海量服务器可以通过平滑将数据包注入网络的突发过程来最大限度地减少数据包丢失。在中等时间尺度上，海量服务器可以通过捆绑LIKE连接来执行聚合拥塞管理，以避免由于流之间的竞争而导致的突发。在更长的时间尺度上，海量服务器可以映射网络中的持久热点，并优化连接调度，以减轻过度使用这些资源的影响。这些目标代表了研究人员打算在这些不同的时间尺度上开展的项目。本提案中概述的研究工作旨在通过推导新的测量和分析技术来实现这些好处，以实现网络状况的诊断，并开发新的服务和协议以实现有效的网络资源管理和控制。在海量服务器的网络测量和诊断方面提出的工作集中于使用被动和主动探测技术来识别不同时间尺度上的网络瓶颈(例如拥塞状况)。要使用的技术范围从使用离散小波变换的多分辨率分析到用于分组丢失估计的最大似然估计器。有了这样的诊断信息，拟议的海量服务器网络资源管理和控制工作侧重于缓解与各种时间尺度上的网络瓶颈(例如，大延迟和抖动、低和不公平的资源利用)相关的问题。要使用的技术范围从开发流量调步协议以减轻亚往返时间尺度上的突发性，到聚合拥塞控制算法以提高网络资源的利用率和减少抖动。拟议工作的一个关键组成部分是实施和原型制作。为此，将通过将所开发的工具和协议整合成三个模块化组件以分发给研究界来展示它们的效用，即：(1)信标：网络测量和诊断工具的集合；(2)收费公路：网络管理和控制协议和服务的集合；以及(3)Backbay：支持将信标和收费公路的功能集成到高性能网络服务器架构中的平台。对本提案中概述的研究目标的追求是及时的。实现这些目标将超越目前旨在支持互联网增长的零碎尝试。为追求这些雄心勃勃的目标而组建的研究小组在了解互联网流量特征方面做出了国家公认的重大贡献，并在软件开发和技术转让方面拥有既定的记录。波士顿大学致力于通过大量的财政和基础设施承诺来支持这个团队，以补充和利用从NSF寻求的支持。

This proposal seeks funding to investigate a number of basic research problems within the general framework of network-aware server protocols and services. Tackling these problems is critical for the continued scalability of the Internet. Mass servers are popular Internet servers which produce a substantial fraction of the traffic flowing through the network. Currently, such Mass servers are tuned to optimize their own performance (e.g. capacity, latency, I/O utilization), while overlooking the opportunity to monitor network conditions and to use that information advantageously. This information could be used both to improve network behavior (e.g. alleviate the problems resulting from congestion) and in further optimizing the performance of servicing requests. Thus, Mass servers are uniquely positioned (1) to observe and diagnose network conditions by tracking the flows that they generate, and (2) to manage and control network resources by better regulating and scheduling the traffic they inject into the network. These goals must be pursued over a wide spectrum of time scales to maximize the beneficial impact that can be achieved. On the shortest time scales, a Mass server can minimize packet loss by smoothing the otherwise bursty process of injecting packets into the network. At medium time scales, a Mass server can perform aggregate congestion management by bundling like connections to avoid the burstiness that results from competition among flows. At even longer time scales, a Mass server can map persistent hotspots in the network and optimize scheduling of connections to mitigate the impact of overusing those resources. These objectives are representative of projects which the researchers intend to undertake on these various time scales. The research work outlined in this proposal aims at achieving these benefits through the derivation of new measurement and analysis techniques to enable diagnosis of network conditions, and the development of new services and protocol to enable efficient network resource management and control. Proposed work in network measurement and diagnosis at Mass servers focuses on the use of passive and active probing techniques for the identification of network bottlenecks (e.g. congestion conditions) along various time scales. The techniques to be used range from multi-resolution analysis using discrete wavelet transforms, to maximum likelihood estimators for packet loss estimation. Empowered with such diagnostic information, the proposed work in network resource management and control at Mass servers focuses on alleviating the problems associated with network bottlenecks (e.g. large delays and jitters, low and unfair resource utilization) along various time scales. The techniques to be used range from the development of traffic pacing protocols to alleviate burstiness at the sub-round-trip-time scale, to aggregate congestion control algorithms for improving utilization of network resources and reducing jitter. A key component of the proposed work is implementation and prototyping. To that end, the utility of the tools and protocols developed will be demonstrated by integrating them into three modular components for distribution to the research community, namely: (1) BEACON: A collection of network measurement and diagnosis tools, (2) TURNPIKE: A collection of network management and control protocols and services, and (3) BACKBAY: A platform that supports the integration of BEACON and TURNPIKE functionality into a high performance web server architecture. The pursuit of the research goals outlined in this proposal is timely. Achieving these goals will leapfrog current piecemeal attempts aiming at supporting Internet growth. The research team assembled to pursue these ambitious goals has made significant, nationally-recognized contributions to the understanding of Internet traffic characteristics and has an established record in software development and technology transfer. Boston University is committed to supporting this team through substantial financial and infrastructural commitments that complement and leverage the support sought from NSF.