计算机工程与应用 ›› 2016, Vol. 52 ›› Issue (23): 6-11.

• 热点与综述 • 上一篇    下一篇

面向可重构系统的负载均衡低能耗调度算法

敬  超   

  1. 1.桂林理工大学 “嵌入式技术与智能信息处理”广西高校重点实验室,广西 桂林 541004
    2.桂林理工大学 信息科学与工程学院,广西 桂林 541004
    3.上海交通大学 电子信息与电气工程学院,上海 200240
  • 出版日期:2016-12-01 发布日期:2016-12-20

Load-balancing based energy-efficient scheduling algorithm for reconfigurable systems

JING Chao   

  1. 1.Guangxi Key Laboratory of Embedded Technology and Intelligent Information Processing, Guilin University of Technology, Guilin, Guangxi 541004, China
    2.School of Information Science and Engineering, Guilin University of Technology, Guilin, Guangxi 541004, China
    3.School of Electronic Information & Electrical Engineering, Shanghai Jiaotong University, Shanghai 200240, China
  • Online:2016-12-01 Published:2016-12-20

摘要: 主要研究了基于多FPGAs部件的可重构系统高能耗问题。首先,对多FPGAs部件可重构系统的特征进行了建模,包括重构端口受限、资源受限及通信开销等建立了问题模型;接着,基于概率论与统计学的离散方差理论,采用负载均衡思想设计和实现了一种低能耗调度算法MLB。它的原理是通过计算各个FPGA部件的总能耗方差来引导负载的均衡分配。最后,通过模拟仿真实验,将提出的MLB算法分别与贪心算法和最新研究MFIT算法进行了比较,结果表明提出的算法复杂度低、运行速度快,不仅多节约了15%的能量,而且缩短了最大完成时间。

关键词: 可重构系统, 多现场可编程门阵列(FPGAs)部件, 负载均衡, 低能耗调度

Abstract: This paper studies the crucial problem of energy-efficiency on multi-FPGA based reconfigurable systems. Firstly, based on the characteristics of limited reconfiguration ports, resources and communication cost on reconfigurable systems, it establishes the problem model. Then, due to the importance of load-balancing for energy reduction, based on the probability theory and statistics, a loading balance algorithm for energy optimization(MLB) is proposed to address the high energy consumption problem. At last, it develops comprehensive trace-driven simulation experiments to evaluate the algorithm, the results show that the proposed algorithm is high efficiency with low-complexity. Compared with Greedy and the latest MFIT, MLB saves 15% energy more than that of two algorithms. Also, MLB shortens the maximum makespan.

Key words: reconfigurable system, multiple Field-Programmable Gate Arrays(FPGAs), load balancing, energy-efficient scheduling