一种基于行为特征的文件检查点优化策略

A Behavior Based File Checkpointing Strategy

检查点机制是高性能计算平台的一项重要特性。它能够在程序运行的某一时刻保存程序的运行状态,并在系统故障后恢复程序状态继续执行。由于文件操作在应用程序中的普遍性,支持文件回卷对于检查点技术来说是十分必要的。文件数据备份可以使文件在回卷后恢复到正常状态,但是开销太大。本文提出了一种基于行为特征的文件检查点优化策略(BBFC),能够提供文件数据的正确恢复,有效保证了程序回卷恢复到上一个检查点时文件状态与进程其它状态保持一致。BBFC对文件行为特征进行分类,并根据这些行为特征采取相应的保存恢复策略,从而在很大概率上减少了检查点间隔需要保存的文件内容,降低了文件检查点的时间、空间开销。它对用户透明,简单易用。

Checkpoint/restart is an important feature of high performance platforms. It saves the state of an exe-cuting program periodically and recovers it after a failure. As many applications involve file operations, supporting file rollbacks is essential for checkpoint/ restart. Backup can restore files to the correct state, but its cost is too high.In this paper we propose a behavior based file checkpointing strategy （ BBFC）, which provides a correct recovery of file data and ensures consistency between file state and other states of a process when a rollback is done by restart-ing the program from the last checkpoint. BBFC classifies details of the file operation behaviors and provides a guidance on what to be saved during file checkpointing according to those behaviors. It dramatically minimizes the overhead of file checkpointing due to the reduction of file data which need to be saved. And it is transparent and easy to use.