Project seeks to balance power, performance in embedded computers

“The computing world is moving from the desktop and workstation to an arena of embedded and wearable computers,” remarked Sandeep Shukla, who recently received a $400,000 grant from the National Science Foundation to help solve one of the major problems in this transition.

Shukla, who joined the Virginia Tech electrical and computer engineering faculty in August 2002, will use his Faculty Early Career Development Program (CAREER) Award to devise a strategy for achieving the optimal balance of power and performance in embedded computer systems.

Embedded computers, he explained, are used in every sphere of modern life. More and more, embedded computers are becoming the brains behind mechanisms that we rely on throughout our everyday lives — wireless devices, cars, automated elevators, climate control systems, traffic signals, and washing machines, to name a few.

“Some experts estimate that each individual in a developed nation may unknowingly use more than 100 embedded computers daily,” Shukla noted. “Embedded computers also constitute the backbone of our complex systems, such as space mission controls, avionics, and weapons systems.”

Most embedded computers are powered by rechargeable batteries. Because space is limited in their host devices, they typically operate on small, low-power batteries. “These computers must function on low power and at the same time offer a level of performance guaranteed by a Quality of Service (QoS) agreement that serves as an industry standard,” Shukla noted. “In certain situations, as in the case of pace-makers, the batteries must last a long time without replacement.”

There are two performance factors critical to embedded computers — speed and quality of service. “If the power supplied by the battery is too low, the computer’s performance is reduced,” Shukla said. “That might affect the speed of a palm pilot, for example, or the sound quality in a hearing device. The question is whether a compromise between performance and power is reasonable for a particular device or application.”

The trend toward arranging embedded computers in a network also has created a need for research into the optimal balance of power and performance. A car, for example, may have a network of 15 to 20 embedded computers on board. “And someday, through global positioning systems technology, your car will not only tell you that it needs an oil change, but will find an auto shop and make an appointment,” Shukla predicted.

Shukla’s goal is to support the current and future uses of embedded computers by developing a power usage strategy that can guarantee maximum performance. This entails analyzing the complex probabilities of when computers will require power and how much power they will use.

“It’s similar to designing a network of traffic lights for a particular traffic pattern,” he explained. “The highway engineer has to study the probabilities of when and where traffic is the heaviest and then set up a network of lights that will allow a maximum flow of traffic.”

To consider a minor example, a usage strategy could be devised for a cell phone that would put the system in the “sleep” mode during times when the probability of usage is low and keep the system in a “ready” mode when incoming and outgoing calls are expected and fast action is required. Such a strategy would reduce power use and increase the life of the battery while optimizing the cell phone’s performance.

Using a probability analysis modeling tool called PRISM, which he worked with at the University of Birmingham in England during the summer of 2002, Shukla will devise usage strategies for a network of wireless computers. Based on analyses of the usage frequencies and probabilities of all the computers in a networked embedded system, he will attempt to create a strategy that will reduce power use while increasing performance.

“Eventually, companies will use probability design in developing embedded computers for everything from small wireless devices to large-scale computer networks,” Shukla said. One company in the United States and a research institute in France already have expressed interest in the outcome of his research.

Shukla also plans to develop graduate and undergraduate courses in embedded computer systems and to support the work of student assistants in FERMAT (Formal Engineering Research with Modeling, Abstraction and Transformation), the new research laboratory he has founded. He is working on two textbooks and is planning an outreach program for the local public schools.

Shukla received his M.S. and Ph.D. in computer science from the State University of New York. He began studying embedded computers while working as an engineer with Verizon and, later, Intel. Before coming to Virginia Tech, he was a member of the research faculty of the Center for Embedded Computer Systems at the University of California at Irvine.

For more information about this research, contact Sandeep Shukla at 540-231-2133 or shukla@vt.edu.

PR CONTACT: Liz Crumbley, 540-231-9772, lcrumb@vt.edu