Now showing 1 - 3 of 3
  • Publication
    Towards Reliable and Accurate Energy Predictive Modelling using Performance Events on Modern Computing Platforms
    (University College Dublin. School of Computer Science, 2020) ;
    Information and Communication Technologies (ICT) systems and devices are forecast to consume up to 50% of global electricity in 2030. Considering the unsustainable future predicted, energy efficiency in ICT is becoming a grand technological challenge and is now a first-class design constraint in all computing settings. Energy efficiency in ICT can be achieved at the hardware level (or system-level) and software level (or application-level). While the mainstream approach is to minimize the energy of the operating environment and is extensively researched, application-level energy optimization is comparatively understudied and forms the focus of this work. The fundamental building block for energy minimization at the application level is an accurate measurement of energy consumption during application execution. There are three popular approaches to providing it: (a) System-level physical measurements using external power meters, (b) Measurements using on-chip power sensors, and (c) Energy predictive models. While the first approach is considered to be the ground truth, it is cost-prohibitive for the energy optimization of applications. The second approach using energy measurements by state-of-the-art on-chip sensors is not recommended for energy optimization of applications due to several issues related to the lack of accurate information in the vendor manuals and poorly reported experimental accuracy. The third approach of energy predictive modelling based on performance monitoring counters (PMCs) as model variables is now the leading method for prediction of energy consumption during application execution. In this thesis, we focus on application-level energy measurement, modelling, and optimization using PMCs. A vast majority of research works propose models where the employed model variables (or PMCs) are selected solely on the basis of their high positive correlation with energy consumption and report prediction accuracies ranging from poor to excellent. There are a few pieces of research that critically examine the inaccuracy of PMC-based models. We present a study to identify the causes of inaccuracies. We formulate a sound theoretical framework to understand the fundamental significance of the model variables with respect to the energy consumption and the causes of inaccuracy or the reported wide variance of the accuracy of the models. We use a model-theoretic approach to formulate the assumed properties of existing energy predictive models in a mathematical form. We extend the formalism by adding properties, heretofore unconsidered, that account for a limited form of energy conservation law. The extended formalism defines our theory of energy of computing. By applying the basic practical implications of the theory, we improve the prediction accuracy of state-of-the-art linear regression models from 31.2% to 18.01%. We demonstrate that use of the state-of-the-art measurement tools for energy optimization may lead to significant losses of energy (up to 84% for applications used in experiments) since they do not take into account the properties of the theory of energy predictive models for computing. Finally, we present the first comprehensive experimental study to compare the energy predictive modelling techniques employing PMCs. We demonstrate that a platform-level and application-level linear regression-based model employing the additive PMCs, irrespective of the applications used for training and testing, performs better in terms of average prediction accuracies.
  • Publication
    A Novel Statistical Learning-Based Methodology for Measuring the Goodness of Energy Profiles of Applications Executing on Multicore Computing Platforms
    Accurate energy profiles are essential to the optimization of parallel applications for energy through workload distribution. Since there are many model-based methods available for efficient construction of energy profiles, we need an approach to measure the goodness of the profiles compared with the ground-truth profile, which is usually built by a time-consuming but reliable method. Correlation coefficient and relative error are two such popular statistical approaches, but they assume that profiles be linear or at least very smooth functions of workload size. This assumption does not hold true in the multicore era. Due to the complex shapes of energy profiles of applications on modern multicore platforms, the statistical methods can often rank inaccurate energy profiles higher than more accurate ones and employing such profiles in the energy optimization loop of an application leads to significant energy losses (up to 54% in our case). In this work, we present the first method specifically designed for goodness measurement of energy profiles. First, it analyses the underlying energy consumption trend of each energy profile and removes the profiles that exhibit a trend different from that of the ground truth. Then, it ranks the remaining energy profiles using the Euclidean distances as a metric. We demonstrate that the proposed method is more accurate than the statistical approaches and can save a significant amount of energy.
      54ScopusĀ© Citations 2
  • Publication
    A Comparative Study of Methods for Measurement of Energy of Computing
    Energy of computing is a serious environmental concern and mitigating it is an important technological challenge. Accurate measurement of energy consumption during an application execution is key to application-level energy minimization techniques. There are three popular approaches to providing it: (a) System-level physical measurements using external power meters; (b) Measurements using on-chip power sensors and (c) Energy predictive models. In this work, we present a comprehensive study comparing the accuracy of state-of-the-art on-chip power sensors and energy predictive models against system-level physical measurements using external power meters, which we consider to be the ground truth. We show that the average error of the dynamic energy profiles obtained using on-chip power sensors can be as high as 73% and the maximum reaches 300% for two scientific applications, matrix-matrix multiplication and 2D fast Fourier transform for a wide range of problem sizes. The applications are executed on three modern Intel multicore CPUs, two Nvidia GPUs and an Intel Xeon Phi accelerator. The average error of the energy predictive models employing performance monitoring counters (PMCs) as predictor variables can be as high as 32% and the maximum reaches 100% for a diverse set of seventeen benchmarks executed on two Intel multicore CPUs (one Haswell and the other Skylake). We also demonstrate that using inaccurate energy measurements provided by on-chip sensors for dynamic energy optimization can result in significant energy losses up to 84%. We show that, owing to the nature of the deviations of the energy measurements provided by on-chip sensors from the ground truth, calibration can not improve the accuracy of the on-chip sensors to an extent that can allow them to be used in optimization of applications for dynamic energy. Finally, we present the lessons learned, our recommendations for the use of on-chip sensors and energy predictive models and future directions.
      310ScopusĀ© Citations 33