Learning-Directed Dynamic Voltage and Frequency Scaling Scheme with Adjustable Performance for Single-Core and...

Area of Science:

• Computer Engineering
• Energy-Efficient Computing
• Machine Learning Applications

Background:

• Dynamic Voltage and Frequency Scaling (DVFS) is crucial for energy conservation in computing systems.
• Traditional DVFS models often rely on complex mathematical formulations.
• Learning-based approaches offer a promising alternative for DVFS prediction.

Purpose of the Study:

• To propose a novel, lightweight learning-directed DVFS method.
• To utilize counter propagation networks for task behavior analysis and DVFS prediction.
• To provide an intelligent mechanism for performance adjustment based on user needs.

Main Methods:

• Implementation of a learning-directed DVFS approach.
• Employment of counter propagation networks for sensing and classifying task behavior.
• Development of an intelligent performance adjustment mechanism.
• Experimental validation on NVIDIA JETSON Tegra K1 and Intel PXA270 platforms.

Main Results:

• The proposed method accurately predicts optimal CPU frequency based on runtime program statistics.
• Achieved significant energy savings, reaching up to 42%.
• Demonstrated effective energy consumption and performance management.

Conclusions:

• The learning-directed DVFS method offers an efficient and accurate solution for energy saving.
• Counter propagation networks are effective for real-time DVFS prediction in embedded systems.
• Users can balance energy efficiency and performance requirements through this intelligent method.