Energy-aware computing with application to data mining algorithms

Abstract

Energy has become an important factor in different aspects of computing technologies, such as reducing server energy for lower financial costs, or mobile device energy for longer battery life. In fact, energy efficiency is the major challenge for Exascale computing and beyond. The goal of our work is to present a unique top-down design methodology for developing energy aware algorithms based on energy profiling. The key idea revolves around identifying and measuring components of code with high energy consumption. Optimizing these software components for performance or energy leads to a major impact on overall computational efficiency. As a result, there are two major contributions in our work: 1. A method for identifying components with high energy consumption in compute-intensive applications. We target operations called kernels, which are frequently used operations in the algorithm. 2. A method for estimating software energy for the identified software components, in particular for kernels and load-store operations. The energy evaluation method involves using isolated code with assembly injection. Furthermore, to ensure reliable results, we use physical energy measurements conducted on specially instrumented circuit boards to provide actual and not just simulated measurements. To evaluate the proposed methods, we conducted three cases studies using well-known DM algorithms: back-propagation (BP) neural network, K-Nearest Neighbors, and Linear Regression. We then conducted a benchmark of energy kernelsfor most commonly used DM algorithms. The results highlight the contributions of kernels and memory energy to total algorithms’ energy. These studies form building blocks for understanding software energy distribution and ultimately energy optimization for DM algorithms.