Dynamic Phase-Based Optimization of Embedded Systems

Abstract

Phase-based optimization specializes system configurations to runtime application requirements in order to achieve optimization goals. Due to potentially large design spaces in configurable systems, one major challenge of phase-based optimization is determining the best configuration for achieving optimization goals without incurring significant optimization overhead during design space exploration. This work proposes phase distance mapping, which uses the correlation between phases and the phases' characteristics to dynamically determine optimal or near-optimal configurations with minimal design space exploration, thereby minimizing optimization overhead.