Optimizing Energy Efficiency for Distributed Dense Matrix Factorizations via Utilizing Algorithmic Characteristics

The pressing demands of improving energy efficiency for high performance scientific computing have motivated a large body of software-controlled hardware solutions that strategically switch hardware components to a low-power state, when the peak performance of the components is not necessary. Although OS level solutions can effectively save energy in a black-box fashion, for applications with random/variable execution patterns, slack prediction can be error-prone and thus the optimal energy efficiency can be blundered away. We propose to utilize algorithmic characteristics to predict slack accurately and thus maximize potential energy savings. Keywords-energy; critical path; algorithmic slack prediction.