PARALLEL DATA LAB 

PDL Abstract

Designing Computer Systems with MEMS-Based Storage

Carnegie Mellon University School of Computer Science Technical Report CMU-CS-00-137, May 2000. Superceded by Proceedings of the 9th International Conference on Architectural Support for Programming Languages and Operating Systems, 2000.

Steven W. Schlosser, John Linwood Griffin, David F. Nagle*, Gregory R. Ganger

Dept. of Electrical and Computer Engineering
School of Computer Science*
Carnegie Mellon University
Pittsburgh, PA 15213

http://www.pdl.cmu.edu/

For decades the RAM-to-disk memory hierarchy access gap has plagued computer architects. An exciting new storage technology based on microelectromechanical systems (MEMS) is poised to fill a large portion of this performance gap, significantly reduce power consumption, and enable many new classes of applications. This research explores the impact MEMS-based storage will have on computer systems. We examine the performance of several device designs under development. Results from five application studies show these devices reduce application I/O stall times by 3-10X and improve overall application performance by 1.6-8.1X. Further, integrating MEMS-based storage as a disk cache achieves a 3.5X performance improvement over a standalone disk drive. Power consumption simulations show that MEMS devices use up to 10X less power than state-of-the-art low-power disk drives. Many of these improvements stem from the fact that average access times for MEMS-based storage are 10X faster than disks and that MEMS devices are able to rapidly move between active and power-down mode. Combined with the differences in the physical behavior of MEMS-based storage, these characteristics create numerous opportunities for restructuring the storage/memory hierarchy.

FULL PAPER, TR VERSION: pdf / postscript
FULL PAPER, CONFERENCE VERSION: pdf / postscript