CAREER: Compiler Optimizations for Low Power

Rutgers University New Brunswick, New Brunswick NJ

Investigators

Abstract

Current low power technology focuses on low power hardware designs and OS-level power management. In contrast, compiler support for low-power is still in its infancy, mainly due to the fact that most classical optimizations for speed and space typically reduce a program's overall power consumption as well, making power as a separate optimization objective redundant. However, recent trends and developments in architecture and OS design have made power dissipation an optimization objective in its own right. These recent developments are (1) more advanced OS-level power management support, (2) availability of low-power wireless communication, and (3) reconfigurable architectures with dynamic voltage scaling capabilities. This project investigates low-power optimizations in the context of the Java programming language. Java has become the language of choice for portable programming across a variety of different architectures, including mobile computers, personal digital assistants (PDA), and embedded systems. Such devices rely on battery power for significant periods of their up-time, if not during the entire time of their operation, making power savings a crucial issue. The new low-power optimizations may trade-off execution time for power savings. They include application-driven power management, location-aware method/task mapping to remote servers, and instruction scheduling for reconfigurable architectures with dynamic voltage scaling.