Computer Engineer Wins Prestigious NSF CAREER Award:
Date: Feb, 9, 2001
Research Will Improve Speed in Digital Data Storage
By: Chris Curran
Phone: (513) 556-1806
Photo by: Colleen Kelley
Archive: Research News
Cincinnati - The National Science Foundation awarded Yiming Hu, an assistant professor in the University of Cincinnati College of Engineering, one of its prestigious CAREER Awards for promising young researchers. The award is worth $230,000 over the next four years.
Hu is in the department of electrical and computer engineering and computer science (ECE&CS), and his research focuses on making computers which can store tremendous amounts of data without slowing down the search for that information.
"Hard disk drives are slow, because they're mechanical devices," explained Hu. "We're trying to build very high performance storage systems with very high capacity that are also very fast." A common approach is to combine disk drives, but Hu said that requires a user to be computer savvy. "Most users don't want to mess with the details."
So, Hu is working to develop "intelligent" systems that would work right on information storage disks to improve performance. Hu said the strategy makes the system perform significantly better, but it remains a very easy system to use.
Right now, Hu is working at the theoretical level. He runs computer simulations to test different ideas very quickly. The most promising techniques will then be used to build an actual prototype in about two years. The goal is to develop a flexible system where hard drives can be added as needed. His group has already tested several new ideas that have shown performance improvement from 70 percent to 10 times faster. Eventually, Hu hopes to develop systems that work 100 times faster than computers available today.
The CAREER Award will also allow Hu to investigate more thoroughly issues related to a computer's CPU cache performance. The CPU is the central processing unit. The cache is the computer's shortcut for finding information you use most often. It works 10-20 times faster than standard memory, but the size of the cache is very limited.
In a paper to be presented at the ACM Joint International Conference on Measurement & Modeling of Computer Systems (SIGMETRICS) in June, Hu and his students will demonstrate that a belief held by many researchers is wrong. "Many researchers believe that caches do not work for multimedia computer applications," he said. "Our data have shown this is not the case" Multimedia applications, which handle audio, video, and pictures, are becoming more and more widely used, so their performance is important.
Hu has already invented one method to improve a computer's CPU cache performance It turns out making the system more efficient also reduces power consumption. "We can cut energy use in half without cutting performance," noted Hu. That's very important for laptops and other mobile computing devices.
Specifically, Hu's method increases the "hit ratio" of caches, and that reduces the overall size of the cache. "It's a more clever way of organizing data in memory," said Hu.
Hu is also collaborating with Professor Soon Chung at Wright State University on a very large data mining project. One of the frustrating things about large data storage system is the inability to find information quickly. Imagine trying to find two related packages in a 100-story building, and you get some idea of the dilemma computer engineers face.
Data mining is a technique that helps identify patterns in data to show relationships that are often unexpected. The applications for business are nearly endless. Insurance companies can make better decisions about who to insure. Stores can reduce their inventories by discovering exactly what their different customers want.
Hu's part of the project involves designing a fast computer storage system that can speed up data mining processes.