Gordon F. Hughes
Associate Director, CMRR
University of California, San Diego
Center for Magnetic Recording Research
9500 Gilman Drive, 0401
La Jolla, CA, 92093-0401
gfhughes@ucsd.edu
(858)534-5317 - Phone
(858)254-2600 - Cell
(858)534-8059 - Fax
Office:
Room 102
iDrives
Intelligent Disk Drives - article from IEEE Spectrum (pdf) Intelligent Disk Drives - presentation by Dr. Hughes (pdf)
The iDrive project studies putting computing tasks inside disk drives, beyond their basic role of data storage and retrieval.
Since their introduction 50 years ago, computer hard disk drives have acted as relatively dumb devices, simply storing and retrieving blocks of data. The first generation of hard disk drives were the size of refrigerators and stored a few megabytes for mainframe computers. Today, standardized 3.5-inch disk drives are nearly universal, from PCs to supercomputers. Drive storage capacity has been increasing at an annual rate of 60% since 1992, and over 100% in recent years-a rate exceeding the 18-month doubling of Moore's law for integrated circuit complexity. Single drives today can store up to 200 gigabytes. Disk drives now have 32-bit internal microprocessors, several megabytes of internal RAM and tens of megabytes of disk capacity reserved for internal drive purposes. This on-board computing power has increased alongside improvements in drive speed and capacity, and has already allowed drives to take on several intelligent tasks beyond basic data storage and retrieval, such as detecting imminent drive failure. This power could be exploited for even more tasks, and the addition of some relatively inexpensive extra processing power would open the door to many functions such as in-drive data searching and encryption, and so shift some of the processing burden from CPUs and networks to disk drives. Also, drives could provide "quality of service" information to their clients to improve the performance of storage networks in large computer systems, such as exact data retrieval times, data error conditions and error correction delays.
This study is sponsored by the UCSD Information Storage Industry Center, a Sloan Foundation Center.

