| | |
Summary: Minimizing Expected Energy Consumption through
Optimal Integration of DVS and DPM
Baoxian Zhao Hakan Aydin
Department of Computer Science
George Mason University
Fairfax, VA 22030
bzhao@gmu.edu, aydin@cs.gmu.edu
ABSTRACT
While Dynamic Voltage Scaling (DVS) and Dynamic Power
Management (DPM) techniques are widely used in real-time
embedded applications, their complex interaction is not fully
understood. In this research effort, we consider the prob-
lem of minimizing the expected energy consumption on set-
tings where the workload is known only probabilistically.
By adopting a system-level power model, we formally show
how the optimal processing frequency can be computed ef-
ficiently for a real-time embedded application that can use
multiple devices during its execution, while still meeting the
timing constraints. Our evaluations indicate that the new
technique provides clear (up to 35%) energy gains over the
|
