Skip to Main Content
The object-based storage system stripes data across large numbers of object-based storage devices (OSDs) to enable parallel data access. Subsequently, the workload presented to the individual OSD will be quite different from that of general purpose file systems. However, many distributed file systems employ general-purpose file systems as their underlying file system. This paper presents a versatile object-based file system (referred to as V-OBFS), an extent and B+tree based file system designed for use in OSDs. The V-OBFS implements flexible disk layout with multiple block size and differentiated free space management for both small objects and large objects. In addition, proposed V-OBFS enables fast object search and mapping between object ID and physical location with an efficient object namespace management. Our experiments show that our user-level implementation of V-OBFS can efficiently prevent file system fragmentation and outperforms Linux Ext2 and Ext3 by a factor of two or three.