Efficient logging controls in journaling file systems for reducing logs and handling fragmented data

Abstract

For the system availability, the journaling technique was applied to most of current well-used file systems. Journaling file systems such as EXT3 [4], ReiserFS [5], XFS [6] and JFS [7] provide fast recovery time from the unclean system shutdown. In these journaling file systems, there are three different journaling modes: writeback, ordered and journaled modes. They have different characteristics with respect to data consistency and run-time performance: The writeback mode does not guarantee the data consistency of any file while this mode provides the best run-time performance among three journaling modes. The ordered mode only guarantees the data consistency of appended files among three types of files, i.e., appended, modified, and deleted files. This mode generally provides a slightly lower performance compared to the writeback mode. On the other hand, the journaled mode provides the best guarantee of data consistency, but, the worst run-time performance in general. In this dissertation, a new journaling mode called the adaptive journaling mode is proposed. In the adaptive mode, the file data blocks are explicitly classified as appended, overwritten and deleted types depending on their properties and different journaling oper ations are applied to each type of file data block. This adaptive mode not only guarantees the same data consistency as the journaled mode but also provides the run-time performance comparable to that of the ordered mode or the writeback mode. Our experiment confirms that the proposed journaling mode is up to twice as fast as the journaled mode of a conventional EXT3 file system. As well, in this dissertation, a technique that efficiently handles fragmented data work- loads is proposed. The technique, called HFD, reduces positioning time of a disk in writing them, and thus, enables fast writebacks. The HFD technique is based on a journaling file system which provides the reliability and a high level of data consistency by handling...