Autoregressive-based online backup scheme for small files

Abstract

Reduce backup window with lower system overheads is a challenge for backup system in massive small files environment. In this article, a new autoregressive-based online backup scheme for small files (ABBSF) is presented to achieve the goal by enhancing the utilization ratio of bandwidth resource in normal times. ABBSF exploits single transfer pattern and package transfer pattern to deal with small files. ABBSF employs single transfer to send one small file to the backup side periodically, and collects the network situation in this procedure. According to these network performance information, it builds autoregressive-based model for predicting available network bandwidth in the next period. Meanwhile, it provides predictive step length control to make more utilization of predictive model for maintaining low system overheads. On the basis of prediction, ABBSF makes appropriate transfer policy for package transfer. Finally, it employs package transfer to pack several files and sent them to the backup side. Experiments show that ABBSF has a good performance of prediction with relatively low system overheads to make full use of network bandwidth when backup small files.