Energy Wall for Exascale Supercomputing

Authors

  • Zhiyuan Wang State Key Laboratory of High Performance Computing, National University of Defense Technology, Changsha, Hunan, 410073
  • Yuhua Tang State Key Laboratory of High Performance Computing, National University of Defense Technology, Changsha, Hunan, 410073
  • Juan Chen State Key Laboratory of High Performance Computing, National University of Defense Technology, Changsha, Hunan, 410073
  • Jingling Xue School of Computer Science and Engineering, University of New South Wales
  • Yun Zhou School of Computer, National University of Defense Technology, Changsha, Hunan, 410073
  • Yong Dong School of Computer, National University of Defense Technology, Changsha, Hunan, 410073

Keywords:

Energy consumption, scalability, exascale computing, energy-efficiency speedup, energy wall

Abstract

"Sustainable development" is one of the major issues in the 21st century. Thus the notions of green computing, green development and so on show up one after another. As the large-scale parallel computing systems develop rapidly, energy consumption of such systems is becoming very huge, especially system performance reaches Petascale (10^15 Flops) or even Exascale (10^18 Flops). The huge energy consumption increases the system temperature, which seriously undermines the stability and reliability, and limits the growth of system size. The effects of energy consumption on scalability become a growing concern. Against the background, this paper proposes the concept of "Energy Wall" to highlight the significance of achieving scalable performance in peta/exascale supercomputing by taking energy consumption into account. We quantify the effect of energy consumption on scalability by building the energy-efficiency speedup model, which integrates computing performance and system energy. We define the energy wall quantitatively, and provide the theorem on the existence of the energy wall, and categorize the large-scale parallel computers according to the energy consumption. In the context of several representative types of HPC applications, we analyze and extrapolate the existence of the energy wall considering three kinds of topologies, 3D-Torus, binary n-cube and Fat tree which provides insights on how to mitigate the energy wall effect in system design and through hardware/software optimization in peta/exascale supercomputing.

Downloads

Download data is not yet available.

Downloads

Published

2017-02-07

How to Cite

Wang, Z., Tang, Y., Chen, J., Xue, J., Zhou, Y., & Dong, Y. (2017). Energy Wall for Exascale Supercomputing. Computing and Informatics, 35(4), 941–962. Retrieved from https://www.cai.sk/ojs/index.php/cai/article/view/1960