ISSN 1000-1239 CN 11-1777/TP

计算机研究与发展 ›› 2019, Vol. 56 ›› Issue (8): 1772-1782.doi: 10.7544/issn1000-1239.2019.20170694

• 系统结构 • 上一篇    下一篇

基于能耗梯度的SSD功率建模方法研究

孙鉴1,2,李战怀2,李强1,张晓2,赵晓南2   

  1. 1(北方民族大学计算机学院 银川 750021);2(西北工业大学计算机学院 西安 710129) (2014132@nun.edu.cn)
  • 出版日期: 2019-08-01
  • 基金资助: 
    国家自然科学基金项目(61402017,61502392,61472323);宁夏自然科学基金项目(2018AAC03254);宁夏高等学校科学技术研究项目(NGY2016160);宁夏重点研发计划项目(2018BEB04002);“计算机应用技术”宁夏自治区重点学科项目;北方民族大学校级科研项目(2017JSKY01)

SSD Power Modeling Method Based on the Gradient of Energy Consumption

Sun Jian1,2, Li Zhanhuai2, Li Qiang1, Zhang Xiao2, Zhao Xiaonan2   

  1. 1(School of Computer Science and Engineering, North Minzu University, Yinchuan 750021);2(School of Computer Science and Engineering, Northwestern Polytechnical of University, Xi’an 710129)
  • Online: 2019-08-01

摘要: 近年来闪存芯片(NANDFLASH)的生产技术获得了长足进步,单位芯片的存储容量及数据吞吐率不断提高.闪存芯片已经在移动终端领域成为主流的存储部件,例如在手机、数码相机、单片机等方面已经有了很广泛的应用.随着闪存成本的降低,其应用范围也逐渐扩展至大规模的数据存储系统中.针对在存储系统中闪存能耗预估准确性不高的问题,提出了一种基于能耗梯度的固态硬盘能耗建模方法,有效提升了固态硬盘(solid state disk, SSD)的能耗预测精度.首先根据SSD内部闪存芯片的层次结构及工作原理对SSD在读写过程中的能耗产生原因进行了分析和建模;其次将SSD内部的交错性及并行性作为建立能耗梯度列表的依据,使用测算结合的方法获得固态硬盘的能耗梯度列表,再根据能耗梯度去预测SSD的实时能耗.该建模方法所用的采集方法不会对系统带来额外的性能开销,适用于在线以及离线的SSD能耗预估.实验证明:与传统的线性能耗模型相比,该建模方法在读写操作的能耗预测精度上都有显著的提高.

关键词: 闪存, 非易失性存储器, 存储系统, 能耗模型, 节能

Abstract: In recent years, flash memory chips (NANDFLASH) production technology has been improved, the storage capacity and data throughput enhances unceasingly. NAND flash SSDs have become the preferred storage device in both consumer electronics and datacenters. Flash has superior random access characteristics to speed up many applications and consumes less power than HDDs. Flash chips has become a mainstream storage components in the field of mobile terminal. But with the reduce of the flash memory cost, its application range is gradually extended to mass data storage system. In view of the lower forecast accuracy for flash drives energy consumption in the storage system. We propose a gradient-based SSD power modeling method that estimates the power consumption of storage workloads, and it effectively enhances the energy consumption prediction precision. This method is based on the hierarchical structure and working principle of NAND flash chips, and analyzes the energy consumption in the process of reading and writing. we build energy consumption gradient list with alternation and parallelism operation, and predict energy consumption of SSD. This kind of modeling method will not bring additional performance overhead to the system. The experimental results show the prediction accuracy of gradient-based modeling method has been improved significantly compared with the traditional linear model.

Key words: flash memory, non-volatile memory, storage system, energy-consuming model, energy efficient

中图分类号: