High-performance I/O using MEMS-based Storage for Scientific Applications
Overview
MEMS-based storage systems are potentially a whole new storage technology
capable of a dramatic decrease in entry cost, access time, volume, mass,
power dissipation, failure rate, and shock sensitivity. The MEMS-based storage
device is intended to fill the large RAM/disk access gap and to create complete
system-on-a-chip solutions for computer systems. It can also be designed to
take the place of traditional storage system using hard disks and/or tapes
to achieve either higher performance or better reliability.
In our research, we are using MEMS-based storage in parallel I/O system
to largely improve parallel I/O performance. The high performance,
distinguished two-dimensional data layout and access method, and the built-in
parallelism make it possible to improve performance for current parallel
I/O system using this new type of storage device, especially for parallel
scientific applications and those reading/writing multi-dimensional arrays in
various access patterns. We create an infrastructure to study the performance impact of MEMS-based storage on modern parallel I/O system and try to design software optimizations for such system to improve the I/O performance for various access patterns discovered in scientific applications.
Research Topics
-
Architecture: Using MEMS-based storage in server-side remote storage, or in client-side local storage
-
Applications with different access patterns:
- Scientific applications (multidimensional arrays w/ different access patterns)
- Commercial database systems
- Out-of-core applications (out-of-core access in both remote storage and local storage)
-
MEMS-based storage as large local buffer/cache
-
MEMS-based storage in fault tolerance for disk arrays
-
File partition and data layout in MEMS-based storage
Infrastructure
We implemented an infrastructure to simulate/emulate parallel I/O.
There are three levels in this infrastructure: client w/ local storage,
parallel file system, I/O server w/ end storage,
as in Figure 1. In current implementation, each client can make I/O request providing arguments (file, offset, len, buffer), while the servers will perform
underlying I/O by sending storage request and providing arguments
(block\#, block-offset, len) to the end storage.
Clients can do caching/buffering of data in their local storage, which
has not been implemented currently.
Figure 1. Parallel I/O Infrastructure Using MEMS-based Storage
In this infrastructure, there are two places that may benefit from MEMS: we can either put MEMS in the remote storage for general purpose performance improvement (maybe more than that) or put it in local storage for caching/buffering optimizations. Above this infrastructure, we can run different kinds of applications with various interesting access patterns, just requiring them to feed their I/O requests to the clients in the form of (file, offset, len, buffer).
People
Alok Choudhary, Jianwei Li, Wei-keng Liao (Northwestern University)
Rob Ross, Rajeev Thakur (Argonne National Laboratory)
Related Links
References
-
Schlosser, S., Griffin, J., Nagle, D., Ganger, G., Filling the Memory Access Gap: A Case for On-Chip Magnetic Storage. Technical Report CMU-CS-99-174, Carnegie Mellon University School of Computer Science, November 1999.
-
Schlosser, S., Griffin, J., Nagle, D., Ganger, G., Designing Computer Systems with MEMS-based Storage. In ASPLOS 2000, November 13-15, 2000.
-
L. Richard Carley, Gregory R. Ganger, and David F. Nagle, MEMS-Based Integrated-Circuit Mass-Storage Systems. in COMMUNICATIONS OF THE ACM November 2000, Vol.43, No.11.
-
Griffin, J., Schlosser, S., Ganger, G., Nagle, D., Operating Systems Management of MEMS-based Storage Devices. In OSDI 2000, October 23-25, 2000.
-
Griffin, J., Schlosser, S., Ganger, G., Nagle, D., Modeling and Performance of MEMS-Based Storage Devices. In Proceedings of SIGMETRICS 2000, June 18-21, 2000. Published as Performance Evaluation Review 28(1):56-65, June 2000.
-
Tara M. Madhyastha, Katherine Pu Yang, Physical Modeling of Probe-Based Storage.In Proceedings of the Eighteenth IEEE Symposium on Mass Storage Systems (April 2001).
-
Z.N.J. Peterson, S.A. Brandt and D.D.E. Long. Data Placement Based on the Seek Time Analysis of a MEMS-based Storage Device. A Work in Progress (WIP) at: the Conference on File and Storage Technologies (FAST), USENIX, 2002.
-
R.C. Burns, R.M. Rees, Z.N.J. Peterson, and D.D.E. Long. Allocation and Data Placement Using Virtual Contiguity. iNIST/SSRC/01-001. 2001.
-
Pu Yang. Modeling Probe-based Storage Devices. Tehcnical report. Department of Computer Science, University of California Santa Cruz, June 2000. Master's thesis.
-
Feng Wang, Scott Brandt. Using MEMS Device as Disk Write Buffer. Department of Computer Science, University of California Santa Cruz. 2001
-
Ying Lin, Power consumption on MEMS-based storage device. Report for cmps290s, December 1, 2001
-
John Linwood Griffin, Jiri Schindler, Steven W. Schlosser, John S. Bucy, Gregory R. Ganger, Timing-accurate Storage Emulation. In Proceedings of the Conference on File and Storage Technologies (FAST), January 28-30, 2002.
-
P. Vettiger, M. Despont, U. Drechsler, U. Dürig, W. Häberle, M. I. Lutwyche, H. E. Rothuizen, R. Stutz, R. Widmer, G. K. Binnig. The 'Millipede'--More than one thousand tips for future AFM data storage. IBM Journal of Research and Development, Directions in information technology, Vol. 44, No. 3, 2000
-
H. J. Mamin, R. P. Ried, B. D. Terris, and D. Rugar, High-Density Data Storage Based on the Atomic Force Microscope. Proc. IEEE 87, No. 6, 1014-1027 (June 1999).
-
B. Hong, Exploring the Usage of MEMS-based Strorage as Metadata Storage and Disk Cache in Storage Hierarchy. Department of Computer Science, University of California Santa Cruz. 2003
-
Steven W. Schlosser, Jiri Schindler, Anastassia Ailamaki, and Gregory R. Ganger, Exposing and Exploiting Internal Parallelism in MEMS-based Storage. Carnegie Mellon University Technical Report CMU-CS-03-125, March 2003.
-
M. Uysal, A. Merchant, and G. Alvarez, Using MEMS-based stroage in disk arrays. Proc. of 2nd USENIX Conference on File and Storage Technologies, April 2003
-
Hailing Yu, Divyakant Agrawal, and Amr El Abbadi, Towards Optimal I/O Scheduling for MEMS-Based Storage. 20 th IEEE/11 th NASA Goddard Conference on Mass Storage Systems and Technologies (MSS'03) April 07 - 10, 2003
San Diego, California
-
H. Yu, D. Agrawal, and A. E. Abbadi, Tabular Placement of Relational Data on MEMS-based Storage Devices. In proceedings of the 29th Conference on Very Large Databases(VLDB), 680-693. September 2003
-
H. Yu, D. Agrawal, and A. E. Abbadi, Declustering two-dimensional datasets over MEMS-based Storage. Computer Science Department, UCSB, 2003
last updated 03/30/2004