Skip to main content
 
HOME FOR INVENTORS FOR INDUSTRY FOR ENTREPRENEURS
 

Details

Project TitleExecution Time Prediction for Energy-Efficient Computer Systems
Track Code7165
Short Description

This technology provides a proactive, automatic predication-guided framework for controllers for hardware accelerators and software applications, enabling greatly improved energy efficiency of computer systems.
   

Abstract

In computing applications and systems, a trade-off exists between decreasing the response time of a software application or hardware task and improving the energy efficiency of the system. As the response time is reduced and performance increases, the energy consumption also increases. However, while a computing task should be completed fast enough to maintain a smooth user experience, improving performance and response time beyond what a user is able to perceive may unnecessarily increase the energy consumption of the system without providing any additional benefit.

  

Techniques have been created to address this trade-off between power consumption and response time; however, most are reactive to changing input variables and cannot respond fast enough to provide a significant benefit. The few proactive or predictive controllers that exist are narrowly tailored for specific applications and are not broadly applicable to a variety of computing tasks.

   

The Cornell prediction-based framework for generating controllers significantly improves the energy efficiency with little to no impact on the user experience. This framework has been developed for software applications and for dedicated hardware accelerators. The controller developed for software applications automatically adapts to changes in environments, such as interference from other applications, and can be easily applied across diverse platforms. The prediction-guided controller for software applications resulted in energy savings of 56% with almost no response time deadlines missed. In addition, the execution time prediction controllers for hardware accelerators have shown 36.4% energy savings while eliminating response time deadline misses.

  

Potential Applications

  • Computer software and hardware systems
  • Mobile computing systems
  • Data centers

   

Advantages

  • Improves energy efficiency of computing systems

   

 
TagsDVFS, computer energy efficiency, mobile computing, run-time prediction
 
Posted DateOct 5, 2016 4:44 PM

Researcher

Name
Gookwon Suh
Daniel Lo
Tao Chen
Taejoon Song

Additional Information

Licensing Contact

Martin Teschl
mt439@cornell.edu
(607) 254-4454

Files

File Name Description
D7165 Technology Brief None Download