| Summary: | 碩士 === 國立暨南國際大學 === 資訊管理學系 === 93 === ABSTRACT
Among the file systems, Log-Structured File System (LFS) is famous for its optimization for write performance. Because it applies append only and non-update-in-place method in data writing, garbage collection is needed to reclaim the space occupied by the obsolete data. However, the cleaning process incurs large overhead and could significantly decrease the performance of file system. Therefore, many researches focus on reducing the cleaning overhead. Traditional cleaning policies do not consider the storage location where the valid data in the selected segments for cleaning should be placed and rewritten to. Whereas, our research considers the storage layout and takes advantage of the chance of data reorganization during garbage collection and data writing to dynamically reorganize the current disk layout, with the aims to reduce the cost of garbage collection and dynamically reorganize the current disk layout to make it an efficient disk layout for successive data accesses.
In this thesis, a new method called Region-based LFS (R-LFS) is proposed, in which disk space is partitioned into several regions and frequently accessed data would be migrated to the central disk in order to reduce disk seek time. Besides, a cleaning policy called Automatic Blocks Reorganization (ABR) is proposed. Based on the segment read frequencies, it builds a minimization of seek time (MST) disk layout, and the current disk layout will be adapted to MST disk layout during garbage collection.
Because R-LFS uses block update frequency to cluster cold and hot data to different regions during data writing and garbage collection and ABR uses segment read frequency to reorganize disk layout during garbage collection to adapt to the MST disk layout, we combine the advantages of R-LFS and ABR to propose a new method called Region-based Automatic Blocks Reorganization (R-ABR). R-ABR uses ABR for selecting the target segment for cleaning and uses R-LFS for writing data into disk.
A trace-driven simulation and a practical implementation on NetBSD/LFS are used in the performance evaluation. Experimental results show that our proposed methods can significantly improve the performance of Log-Structured File System.
Keyword: Log-Structured File System, Garbage Collection, Data Reorganization
|
|---|
