容错系统中文件状态的保存与恢复算法

在机群计算环境中实现容错是人们日益关心的热点。许多著名的机群计算环境都使用检查点实现了容错的功能。但目前的检查点算法在使程序卷回执行的同时，不能相应地恢复文件系统的状态，因而对应用程序访问文件系统有较多的限制。本文在原子操作和并发控制的基础上，提出了能够恢复文件系统状态的ＳＣＲ算法，进一步发展了文件系统可恢复性的概念，ＳＣＲ算法与检查点机制结合使用，便可支持分布式应用程序在容错运行时对文件系统的任

AN ALGORITHM OF SAVING AND RESTORING A STATE OF A FILE SYSTEM

Fault-tolerance is very important in cluster computing. Many famous cluster computing systems have implemented fault-tolerance by using checkpoint/ restart mechanism. But existent checkpoint/restart algorithms can not restore a state of a file system when roll-backing the running of a program, so there are many restrictions to file accesses in existent fault-tolerance systems. SCR algorithm, an algorithm based on atomic operation and consistent schedule, which can restore a state of file systems, is presented in this paper. In SCR algorithm, file operations used by user applications are classified into idempotent operations and non-idempotent operations. A n0n-idempotent operation modifies a file system's states, and an idempotent operation does not. SCR algorithm dynamically follows the tracks of a program's running, and logs each non-idempotent operation used by the program and the related information that can restore the operation in disk.When checkpointing roll-backing the program's running, SCR algorithm will revert the file system's states to the last checkpoint cut. By using SCR algorithm, users are allowed to use any flle operation in their programs.