CSR-PDOS: ISG: Collaborative Research: Building distributed, wide-area applications using WheelFS

分布在广域网上数千台机器上的大型应用程序很难构建和维护。为了实现跨机器共享数据，许多应用程序使用专门的存储和数据传输工具，如分布式哈希表、SCP或GridFTP。虽然文件系统已经成功地成为集群应用程序的通用构建块，但文件系统是否能为广域分布式应用程序提供类似的好处仍不清楚。本研究描述了一种新型的广域文件系统WheelFS，它允许分布式应用程序开发人员使用通用文件系统接口在广域机器之间轻松地存储和共享应用程序数据。两种新的方法使WheelFS对分布式应用程序的使用具有吸引力。首先，WheelFS为应用程序开发人员提供语义提示，以在单个文件和目录的粒度上表达在容错能力、数据一致性和文件系统性能之间所需的权衡。其次，WheelFS通过将应用程序数据写入本地磁盘并尽可能从附近机器读取缓存副本来优化广域数据传输。该项目通过构建多个分布式应用程序的经验展示了WheelFS的无用之处，这些应用程序包括协作Web缓存、数据密集型网格应用程序、分布式数字图书馆和PlanetLab测量实用程序。

Large-scale applications distributed across thousands of machines over the wide area network are difficult to build and maintain. To enable sharing of data across machines, many applications use specialized storage and data transfer tools such as a DHT, scp or GridFTP. Although file systems are successful in becoming a common building block for cluster applications, it remains unclear if file systems could provide similar benefits to wide area distributed applications. This research describes a novel wide-area file system, WheelFS, that allows distributed application developers to use a generic file system interface to store and share application data easily among wide-area machines. Two new approaches make WheelFS attractive for use by distributed applications. First, WheelFS provides semantic cues for application developers to express desired tradeoffs among failure resilience, data consistency and file system performance at the granularity of individual files and directories.Second, WheelFS optimizes wide area data transfer by writing application data to local disks and reading a cached copy from a nearby machine whenever possible.This project demonstrates the uselessness of WheelFS via the experience of building a number of distributed applications, such as a cooperative web cache, a data-intensive Grid application, a distributed digital library and aPlanetLab measurement utility.