Collaborative Research: Software Architectures for Distributed Web Services Based on Peer-to-Peer Technologies

当今IT研究人员面临的最紧迫问题之一是如何提供分布式服务器基础设施来适应呈指数级增长的网络访问。本项目考虑嵌入在物理网络之上的虚拟网络中由地理上分布的服务器、客户端缓存、代理缓存和分布式搜索组成的Web服务。开发了一套基于对等(P2P)算法的软件技术，以在现有硬件/软件资源的基础上实现最优的整体性能和可靠性。该项目由加州大学河滨分校和萨福克大学联合开展。惠普实验室的研究人员作为高级人员参与其中，本项目考虑的Web服务器是由分布在不同地理位置的、基于Web服务不同功能和应用的网络中的异构节点组成的。在考虑到互联网系统的拓扑属性的情况下，开发集群技术以创建合适的P2P结构。在群集内部，为Web服务确定了可扩展、高效和高性能的覆盖网络体系结构。遵循分布式共享内存(DSM)的方法，在虚拟网络中开发了高效的协作代理缓存放置和管理机制。将代理服务器与由浏览器(客户)缓存组成的P2P集群相结合，通过嵌入树结构和目录方案，在代理级设计了一种协作缓存策略。使用PlanetLab分布式环境进行了广泛的软件实施、仿真、测试和评估。

One of the most pressing issues faced by today's IT researchers is how to provide a distributed server infrastructure to accommodate the exponentially increasing web accesses. This project considers web services consisting of geographically distributed servers, client caching, proxy caching, and distributed searching in a virtual network, embedded on top of a physical network. A set of software techniques is developed based on peer-to-peer (P2P) algorithms to achieve optimal overall performance and reliability with existing hardware/software resources. The project is carried out jointly at University of California, Riverside and Suffolk University. Researchers from HP Laboratories are involved as senior personnel.The web server, considered in this project, consists of geographically distributed heterogeneous nodes in a network based on different functionalities and applications of the web service. Clustering techniques are developed to create suitable P2P structures keeping in mind the topological properties of the Internet systems. Inside the cluster, scalable, efficient and high performance overlay network architectures are identified for web services. Efficient cooperative proxy cache placement and management mechanisms are developed in the virtual network following distributed shared memory (DSM) approaches. Proxy servers are combined with P2P clusters consisting of browser (client) caches, and a cooperative caching strategy is designed at the proxy level by embedding tree structures and directory schemes. Extensive software implementation, simulation, testing and evaluation are carried out using the PlanetLab distributed environment.