Proceedings of the 3rd USENIX Conference on File and Storage Technologies (FAST '04). San Francisco, CA. March 31, 2004. Supercedes Carnegie Mellon University Parallel Data Lab Technical Report CMU-PDL-03-102, December, 2003.
Steven W. Schlosser, Gregory R. Ganger
Parallel Data Laboratory
Electrical and Computer Engineering
Carnegie Mellon University
Pittsburgh, PA 15213
MEMS-based storage devices are a new technology that is significantly different from both disk drives and semiconductor memories. These differences motivate the question of whether they need new abstractions to be utilized by systems, or if existing abstractions will work well. This paper addresses this question by examining the fundamental reasons that the abstraction works for existing systems, and by showing that these reasons hold for MEMS-based storage. This result is borne out through several case studies of proposed roles MEMS-based storage devices may take in future systems, and potential policies that may be used to tailor systems’ access to MEMS-based storage. We argue that when considering the use of MEMS-based storage in systems, their performance should be compared to that of a hypothetical disk drive that matches the speed of a MEMS-based storage device. We discuss exceptional workloads that can use specific features of MEMS-based storage devices and that may require extensions to current abstractions. Also, we consider the ramifications of the assumptions that are made in today’s models of MEMS-based storage devices.
KEYWORDS: MEMS-based storage, storage interfaces, request scheduling, database.
FULL PAPER: pdf / postscript
ORIGINAL TR VERSION OF THIS PAPER: pdf / postscript