Tape storage solutions: meeting growing data demands
David Du, Xianbo Zhang · 2006
The exponential data growth caused by content-rich applications and new data compliance regulations has led to an increased demand for tape storage due to tape's low cost per GB and long shelf-life. However, tape technology suffers from several disadvantages: it provides a hard to use SCSI interface; a tape drive cannot be easily shared and needs to operate in streaming mode to obtain its full performance potential; backup practices that include staging disks increase the cost and complexity of the backup process; and a single tape drive's speed may not be high enough for some applications. This dissertation develops new tape-based solutions to these problems, allowing high capacity, high performance data storage that takes full advantage of the capacity and streaming speed of modern tape technology and meets specific requirements from different users. First, HPTFS, a High Performance Tape File System, is designed and implemented to mount a tape as a file system and enable tape write sharing among multiple users by transparently interleaving user data streams to tape. The developed file system simplifies the tape access and increases tape drive write speed with data buffering and interleaving. HPTFS eliminates costly staging disks in its data path. Next, ITStorage, an Internet Tape Storage system, is designed and prototyped to provide a, cheap, infinite online data storage system to backup data from personal users, small businesses, and enterprise remote offices which are not as well protected as enterprise data centers. With EHPTFS (Extended HPTFS) as the technology core, data interleaving and tape switches are transparent to user applications while tape metadata and media information are consolidated in a commodity relational database. Finally, an object placement scheme for parallel tape storage systems is proposed to address the needs of high performance computing where a huge amount of data transfer between disk storage and tape storage is common. It is important to increase the speed of reading data from tape to disk to increase the utilization of high performance computing resources and minimize system disaster recovery time. The effectiveness of the proposed scheme to increase aggregate data retrieval bandwidth is studied through simulation.