CSR-PDOS: Fast, Consistent Distributed File Systems through Operating System Speculation

CSR-PDOS：通过操作系统推测实现快速、一致的分布式文件系统

• 批准号: 0509093
• 负责人: Jason Flinn
• 金额: $ 60万
• 依托单位: Regents of the University of Michigan - Ann Arbor
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2005
• 资助国家: 美国
• 起止时间: 2005-09-01 至 2009-08-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=0509093&HistoricalAwards=false
• 关键词: CSR; PDOS; Fast; Consistent; Distributed

This project is investigating how generic support for speculativeexecution in a commodity operating system can improve the performanceand consistency of remote data access. Rather than block applicationsduring a remote file system operation, the operating systemcheckpoints the state of the calling process, predicts the result ofthe remote operation, and speculatively resumes execution based uponthe predicted results. If the speculation proves correct, thecheckpoint is discarded; if the speculation is incorrect, the callingprocess is restored to the checkpoint and the operation retried. Theoperating system guarantees safe execution by preventing speculativeprocesses from externalizing output until the speculations upon whichthat output depends have proven correct. By tracking causaldependencies created through inter-process communication, theoperating system allows multiple processes to safely share speculativestate. In addition to file systems, this project is exploring uses ofspeculation in diverse areas, including cyber-foraging, remotedesktops, and software installation.The major impact of this project lies in improving the performance,consistency guarantees, and safety of current distributed filesystems. The project is developing speculative versions of NFS, AFSand other distributed file systems that perform substantially betterthan current versions of these file systems. At the same time, theseversions provide stronger consistency and safety guarantees toapplications. The software artifacts developed as part of thisproject, including a modifed version of Linux that supportsspeculation, are being released to the open-source community.Additionally, educational initiatives are helping train the nextgeneration of systems software developers.

该项目正在研究在商用操作系统中对推测性执行的通用支持如何提高远程数据访问的性能和一致性。操作系统不是在远程文件系统操作期间阻止应用程序，而是检查调用进程的状态，预测远程操作的结果，并基于预测的结果推测性地恢复执行。如果推测被证明是正确的，则丢弃检查点；如果推测不正确，则将调用过程恢复到检查点并重试操作。操作系统通过防止投机进程将输出外部化来保证安全执行，直到输出所依赖的投机被证明是正确的。通过跟踪通过进程间通信创建的原因依赖关系，操作系统允许多个进程安全地共享推测数据。除了文件系统，该项目还在探索投机在不同领域的应用，包括网络寻址、远程桌面和软件安装。该项目的主要影响在于提高当前分布式文件系统的性能、一致性保证和安全性。该项目正在开发推测版本的NFS、AFS和其他分布式文件系统，这些文件系统的性能远远好于这些文件系统的当前版本。同时，这些版本为应用程序提供了更强的一致性和安全性保证。作为这个项目的一部分开发的软件制品，包括支持猜测的Linux的修改版本，正在向开放源码社区发布。此外，教育倡议正在帮助培训下一代系统软件开发人员。