Online frequency-based performance and power estimation for clustered multi-processor systems. (March 2021)

• Record Type: Journal Article
• Authors: Kundan, Shivam; Spantidi, Ourania; Anagnostopoulos, Iraklis
• Is Part Of: Computers & electrical engineering. Volume 90(2021)
• Publication Date: 2021-03
• Subjects: Power-efficient scheduling -- Performance per watt -- Chip multi-processors -- Performance-aware scheduling -- Neural networks; Computer engineering -- Periodicals; Electrical engineering -- Periodicals; Electrical engineering -- Data processing -- Periodicals; Ordinateurs -- Conception et construction -- Périodiques; Électrotechnique -- Périodiques; Électrotechnique -- Informatique -- Périodiques; Computer engineering; Electrical engineering; Electrical engineering -- Data processing; Periodicals; Electronic journals; 621.302854
• DOI: 10.1016/j.compeleceng.2021.106971
• Languages: English
• ISSNs: 0045-7906

Abstract

Abstract: Modern Chip Multi-Processors (CMPs) are required to be increasingly power efficient while also offering higher performance and lower costs. A combination of Dynamic Voltage–Frequency Scaling (DVFS) and sophisticated resource-aware scheduling is needed to address the underlying problem of maximizing performance-per-Watt of CMP architectures. In this paper, we propose a methodology to predict the power consumption and performance for groups of concurrently executing applications at all available frequencies of a CMP. The methodology uses a combination of hardware-based application profiling, contention-aware scheduling, and artificial neural networks. Experimental results on an Odroid-XU3 board demonstrate an increase in average performance per Watt of 30.5% (A15 cluster) and 11.4% (A7 cluster) over Linux's Completely Fair Scheduler (CFS) and power governors. In addition, our methodology outperforms three state-of-the-art resource managers, yielding the highest performance per Watt in all evaluated use cases. Graphical abstract: Highlights: Chip Multicore Processor's power consumption increases exponentially with frequency. However, application performance scales only logarithmically or linearly. We predict the required frequency for specific performance-to-power ratios. Performance-per-Watt improved by 11.4% to 30.5% compared to Linux.