Intelligent Disks are Not the Answer

David J. DeWitt
Computer Sciences Department
University of Wisconsin

Friday, January 25, 2002, 4:30 PM
Volen 106, Brandeis University

(preceeded by a wine and cheese reception at 3:30 pm)

Abstract:

Over the past 20 years CPU performance, disk capacities, and memory sizes have all increased by a factor of 1000. Over the same time period, disk transfer rates have only increased by a factor of 10. While some individuals have advocated "smart disks" as a solution, we think running SQL in the disk controller is going to accomplish very little, if anything at all.

After debunking the novelty of "smart" disks and claims that disks are getting faster, this talk will describe N**3, a new database system project at Wisconsin that is exploring a number of alternative techniques for reducing the I/O bottleneck that is the result of ever slower disk drives. We will describe these techniques and present results indicating their impact on system performance using queries from the TPC-H benchmark.

Speaker Bio:

David J. DeWitt, is a Professor and Chair in the Computer Sciences Department at the University of Wisconsin. His research interests include parallel database systems, database system performance and benchmarking, and distributed database systems. Recently completed research projects include the Paradise database system, a parallel database system designed to support large geographic and image data sets. He is currently involved in the design and implementation of Niagara, a distributed database system for querying distributed repositories of XML documents on the Internet.

Maintained by Dina Goldin dqg AT cse.uconn.edu
Last updated on 1/17/02