PYI: The Sharing of Data in Distributed Environments

本文研究的重点是分布式环境下的数据共享问题。Andrew文件系统解决了在大型分布式环境中提供安全、位置透明的名称空间的问题。该系统现已投入使用，并在CMU校园和其他地方广泛使用。随着对这类系统的依赖增加，能够随时访问数据变得越来越重要。由于服务器和网络故障而导致的数据不可访问在诸如Andrew的系统中造成相当大的不便。在多个站点复制数据是最近理论研究的主题。该项目的目标是在CODA的设计和实现中使用这些算法，CODA是一个对服务器和网络故障具有弹性的分布式文件系统。在正常运行时，该系统的性能不应比同等的非弹性分布式文件系统差。由于复制涉及到向多个服务器传输数据，因此并行网络协议的开发是这项工作的重要组成部分。对替代复制算法的评估、处理网络分区的技术的开发以及对这种系统的分析和建模是该项目的必要部分。

The focus of this research is the sharing of data in distributed environments. The Andrew File System addressed the issue of providing a secure, location transparent name space in a large distributed environment. This system is now operational and widely used on the CMU campus and elsewhere. As dependence on such systems grows, it becomes increasingly important to be able to access data at all times. The inaccessibility of data due to server and network failures causes considerable inconvenience in a system such as Andrew. Replication of data at multiple sites is a technique that has been the subject of much recent theoretical investigation. The project's goal is to use these algorithms in the design and implementation of CODA, a distributed file system that is resilient to server and network failures. In normal operation, the performance of this system should be no worse than that of an equivalent non-resilient distributed file system. Since replication involves data transfer to multiple servers, the development of parallel network protocols is an essential part of this effort. The evaluation of alternative replication algorithms, the development of techniques to handle network partitions, and the analysis and modeling of such a system are necessary parts of the project.