Proceedings of the 1994 International Symposium on Computer Architecture, April 1994, pp. 234-244.
Ann L. Drapeau, Ken Shirriff, Edward K. Lee, John H. Hartman, Ethan L. Miller, Srinivasan Seshan, Randy H. Katz, Ken Lutz, David A. Patterson, Peter M. Chen* and Garth A. Gibson**
UC Berkeley
Berkeley, CA 94720
*Computer Sciencs and Engineering Dept.
University of Michigan
Ann Arbor, MI 48109
**School of Computer Science
Carnegie Mellon University
Pittsburgh, PA 15213
In 1989, the RAID group at UC Berkeley built a prototype disk array called RAID-I. The bandwidth achieved by RAID-I was severely limited by the memory sytem bandwidth limitations of the disk array's host workstation. As a result, most of the bandwidth available from the disks could not be delivered to clients of the disk array file server. We designed our second prototype, RAID-II, to deliver as much of the disk array bandwidth as possible to file sever clients. A custom built circuit board disk array controller, called the XBUS board, connects the disks and the high-speed network directly, allowing data for large requests to bypass the server workstation. A single workstation may control several XBUS boards for increased bandwidth. RAID-II runs the Log-Structrued File System (LFS) to optimize the performance of the disk array for bandwidth intensive applications.
The RAID-II hardware with a single XBUS controller board delivers 20 megabytes/second of I/O between the disks and high-speed networks. This performance is an order of magnitude better than our first prototype, but somewhat lower than our performance goals, because of lower than expected performance of the commercial disk controller boards and our disk system interfaces. A preliminary implementation of LFS delivers 13.4 megabytes/second to the clients.
FULL PAPER: pdf / postscript