The research activities in the lab are partially funded by
NSF CISE CAREER award No. 9985050
"Compiler Optimizations for Low Power" (PI: Ulrich Kremer).
Any opinions, findings, and
conclusions or recommendations expressed in material related to this
project do not necessarily reflect the views of the National
Science Foundation.
Jerry Hom , Department of Computer Science, Rutgers University
Chunling Hu , Department of Computer Science, Rutgers University
Chung-Hsing Hsu, PhD. 2003, Los Alamos National Laboratory, New Mexico
Yang Ni, PhD. 2006 , Intel, Santa Clara, California
Power and Energy Management is a crucial enabling technology for future high-performance desk-top systems, handheld computers, and sensor devices. Power is the rate at which energy is consumed. The more power a device needs, the more heat it dissipates. This heat has to be removed from system components and computers in order to guarantee their operability. Cooling technologies based on airflows have already reached their limits for advanced workstations and servers. Removing heat from machine rooms are a substantial strain on any computing centers budget and the power supply grid.
Energy optimizations typically target battery-operated mobile devices in order to prolong battery life. Reducing the energy consumption of a devices can increase its uptime while disconnected from a steady power supply, or provide the same uptime, but with a lower capacity battery. The latter criterion is important in contexts where the weight and/or size of a device is crucial since batteries represent a significant fraction of a systems weight and volume.
We will investigate compiler-time power and energy optimizations that will be complementary to current hardware and OS techniques. Compilers have the advantage that they can analyze whole program behavior, and reshape this behavior if considered profitable for a given optimization objective. Hardware or OS techniques typically use a window of past program behavior in order to predict future behavior. Code reshaping is only possible within a small window and at a low level of program abstraction.
The following four power and energy management strategies will be investigated for two optimization criteria, (1) minimization of peak power dissipation and (2) minimization of overall energy consumption. The goal of our compilation techniques is to improve the performance of a given program under a single optimization criterion, or both optimization criteria.
We will develop novel compile-time analyses and
compiler transformations to achieve our optimization goals. A
prototype implementation based on the SUIF compiler infrastructure
will be used to validate the efficiency and effectiveness of our
optimizations. Benchmark programs will be taken from a variety of
sources, including computational, multi-media, and image processing
benchmarks.
Last updated by Uli Kremer at 1:50pm on March 21, 2006