Software optimization for performance, energy, and thermal distribution: Initial case studies

Abstract

As an initial step in our Green Software research, this paper investigates whether software optimization at the application level can help achieve higher energy efficiency and better thermal behavior. We use both direct measurements and modeling to quantify power, energy and temperature for a given software method. The infrastructure includes a new power estimator for multicore systems developed by regressing measurements from a custom-designed suite of microbenchmarks. Using our evaluation methodology on a real-life multicore system, we explore two case studies. In the first one, we use software tuning for improving the scalability and energy-efficiency of a parallel application. The second case study explores the effect of temperature optimization on system-level energy consumption.