2018/01/30 by Vi Ngoc-Nha Tran, Tran, Vi Ngoc-Nha, Tommy Oines +5
Computer Science · Engineering · #Advanced Data Storage Technologies #Cloud Computing and Resource Management #Distributed #FOS: Computer and information sciences #Green IT and Sustainability #Parallel #Parallel Computing and Optimization Techniques #and Cluster Computing (cs.DC)
paper · pdf · doi:10.48550/arxiv.1801.10263
openalex publication_date 2018/01/30 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Significant efforts have been devoted to choosing the best configuration of a\ncomputing system to run an application energy efficiently. However, available\ntuning approaches mainly focus on homogeneous systems and are inextensible for\nheterogeneous systems which include several components (e.g., CPUs, GPUs) with\ndifferent architectures. This study proposes a holistic tuning approach called\nREOH using probabilistic network to predict the most energy-efficient\nconfiguration (i.e., which platform and its setting) of a heterogeneous system\nfor running a given application. Based on the computation and communication\npatterns from Berkeley dwarfs, we conduct experiments to devise the training\nset including 7074 data samples covering varying application patterns and\ncharacteristics. Validating the REOH approach on heterogeneous systems\nincluding CPUs and GPUs shows that the energy consumption by the REOH approach\nis close to the optimal energy consumption by the Brute Force approach while\nsaving 17% of sampling runs compared to the previous (homogeneous) approach\nusing probabilistic network. Based on the REOH approach, we develop an\nopen-source energy-optimizing runtime framework for selecting an energy\nefficient configuration of a heterogeneous system for a given application at\nruntime.\n