File system performance and transaction support

作者: Margo Ilene Seltzer

DOI:

关键词: Unix file typesVersioning file systemFile system fragmentationFile systemComputer fileComputer scienceStub fileSelf-certifying File SystemDevice fileOperating system

摘要: This thesis considers two related issues: the impact of disk layout on file system throughput and integration transection support in systems. Historic designs have optimized for reading, as read was I/O performance bottleneck. Since increasing main-memory cache sizes effectively reduce traffic (BAKER91), write has become bottleneck (OUST89). presents both simulation implementation analysis read-optimized write-optimized systems. An example a with writing is log-structured system, where writes are bundled written sequentially. Empirical evidence (ROSE90), (ROSE91), (ROSE92) indicates that provides superior equivalent to traditional systems. analyzes evaluates presented isolating some critical issues its design. Additionally, modified design addressing these evaluated. Log-structured systems also offer potential transaction processing into system. Because use logging techniques store files, incorporating mechanisms natural extension. design, implementation, user-level management embedded optimize system. This shows simple, can attain nearly 100% bandwidth when I/Os large or sequential. The improved LFS discussed only attainable garbage collection overhead small, all workloads examined, comparable On steady stream random issued, reduces by 35% 40%.

参考文章(59)
David B. Golub, Avadis Tevanian, Michael J. Accetta, William J. Bolosky, Richard F. Rashid, Robert V. Baron, Michael Young, Mach: A New Kernel Foundation for UNIX Development. USENIX Summer. pp. 93- 113 ,(1986)
Steve R. Kleiman, Vnodes: An Architecture for Multiple File System Types in Sun UNIX. USENIX Summer. pp. 238- 247 ,(1986)
Margo Seltzer, Michael Olson, LIBTP: Portable, Modular Transactions for UNIX University of California at Berkeley. ,(1992)
John K. Ousterhout, Mendel Rosenblum, The LFS Storage Manager. USENIX Summer. pp. 315- 324 ,(1990)
Nathan Goodman, Philip A. Bernstein, Timestamp-based algorithms for concurrency control in distributed database systems very large data bases. pp. 285- 300 ,(1980)
Bruce W. Leverett, Owen T. Anderson, W. Anthony Mason, Craig Everhart, Edward R. Zayas, Shu-Tsui Tu, Vasilis Apostolides, Michael L. Kazar, Sailesh Chutani, Beth A. Bottos, DEcorum File System Architectural Overview. USENIX Summer. pp. 151- 164 ,(1990)
Marshall K McKusick, Michael J Karels, Keith Bostic, None, A Pageable Memory Based Filesystem. USENIX Summer. pp. 137- 144 ,(1990)
Omri Serlin, Hector Garcia-Molina, Stefano Ceri, Ed Lazowska, Harald Sammer, Mike Stonebraker, Alfred Spector, Jim Gray, Mark Brown, Rick Catell, Bob Good, Kent Trieber, Tony Lukes, Bob Jolls, Pete Homan, Tim Chou, Dave DeWitt, John Nauman, Mike Pong, Peter Weinberger, Dina Bitton, Dieter Gawlick, Andreas Reuter, A measure of transaction processing power Datamation archive. ,vol. 31, pp. 112- 118 ,(1985)
Carolyn Turbyfill, David J. DeWitt, Dina Bitton, Benchmarking Database Systems A Systematic Approach very large data bases. pp. 8- 19 ,(1983)
Michael Stonebraker, Akhil Kumar, Performance Evaluation of an Operating System Transaction Manager very large data bases. pp. 473- 481 ,(1987)