Program phase and runtime distribution-aware online DVFS for combined Vdd/Vbb scaling

• DOI: 10.5555/1874620.1874720
• Publication Date: 2009-04-20
• Authors: Jungsoo Kim; Sungjoo Yoo; Chong‐Min Kyung
• Keywords: Benchmark (surveying); Frequency scaling; Runtime system

Complex software programs are mostly characterized by phase behavior and runtime distributions. Due to the dynamism of two characteristics, it is not efficient make workload predictions during design-time. In our work, we present a novel online DVFS method that exploits both distribution in combined Vdd/Vbb scaling. The presented performs bi-modal analysis distribution, then distribution-aware prediction based on analysis. order minimize overhead sophisticated method, table lookups pre-characterized data without compromising quality energy reduction. It also offers new concept program suitable for DVFS. Experiments show effectiveness case H.264 decoder with sets long-term scenarios consisting total 4655 frames. 6.6% ~ 33.5% reduction consumption compared existing offline solutions.