The design and performance analysis of distributed and parallel computer systems

互联网中的一个关键挑战是向大量不同类型的客户端提供高效的内容传递，例如网页、视频文件和软件包。在无线传感器网络的环境中发现了相反的问题，其中潜在的大量互连的、功率有限的计算设备嵌入到物理环境中。这里的关键挑战是高效地收集和从网络中提取在这些设备上捕获的数据。建议的研究涉及大规模系统中内容传递和数据收集的协议的设计和性能评估，并可分为三个主要项目。据介绍，第一个项目涉及Internet中的内容交付，特别是许多客户端同时请求的大型、高人气文件的按需交付。特别感兴趣的是可扩展的协议，该协议利用多个服务器上的文件复制，从拥有(全部或部分)所请求文件的其他客户端检索数据，以及以从多个源向客户端并发交付文件数据的形式的并行性。他说，第二个项目涉及大规模传感器网络中的数据收集，特别是在数据通过网络转发到采集点时用于网内聚合的协议(S)。特别感兴趣的是异步协议，它确定何时在每个节点自适应地形成聚合，而不是以固定的周期。这些问题包括如何最好地构建聚合树、丢包恢复、适应节点故障和速率控制。最后，第三个项目涉及性能评估方法的进步，包括测量研究和新的分析建模技术的开发，特别是内容交付和传感器网络的应用领域。从这三个项目的贡献来看，这三个项目应显著提高我们对大型系统中内容交付和数据收集/聚合技术的理解和量化评估的能力。

A key challenge in the Internet is that of providing efficient content delivery, such as of web pages, video files, and software packages for example, to large numbers of heterogeneous clients.  A converse problem is found in the context of wireless sensor networks, in which potentially large numbers of interconnected, power-limited computing devices are embedded in the physical environment.  Here a key challenge is that of efficiently collecting and extracting from the network the data that is captured at these devices.  The proposed research concerns the design and performance evaluation of protocols for content delivery and for data collection in large-scale systems, and can be divided into three main projects.      The first project concerns content delivery in the Internet, specifically on-demand delivery of large, highly popular files that many clients concurrently request.  Of particular interest are scalable protocols that make use of file replication at multiple servers,  retrieval of data from other clients that have (all or a portion of) the requested file, and parallelism in the form of concurrent delivery of file data to a client from multiple sources.      The second project concerns data collection in large-scale sensor networks, specifically protocols for in-network aggregation of data as it is being forwarded across the network to the collection point(s).  Of particular interest are asynchronous protocols that determine when to form aggregates at each node adaptively rather than with fixed periodicity.  Issues include how to best construct aggregation trees, packet loss recovery, adapting to node failure, and rate control.      Finally, the third project concerns the advancement of performance evaluation methodologies, including both measurement studies and the development of new analytic modelling techniques, with particular emphasis on the application domains of content delivery and sensor networks.           Contributions from these three projects should significantly improve our understanding of, and ability to quantitatively assess, techniques for content delivery and data collection/aggregation in large-scale systems.