基于温度模糊控制的多核处理器任务调度研究  

Task scheduling for multi-core processors based on fuzzy control

作　　者：纪琮[1] 祝永新[1] 夏亮[1] 

机构地区：[1]上海交通大学微电子学院,上海200240

出　　处：《电子测量技术》2009年第1期32-35,共4页Electronic Measurement Technology

基　　金：上海市人事局及上海市科学技术委员会浦江人才计划(07pj14061)

摘　　要：多核处理器极高的热功耗,对其使用寿命和稳定性造成严重影响。本文采用模糊控制算法,通过SIMULINK模糊仿真分析以及在P4双核处理器上将该算法嵌入至Linux内核调度器,实现任务调度,达到了降低处理器峰值温度的目的。随机调度算法与模糊控制算法的对比实验数据显示,处理器温度超限比例从77%下降到64%,处理器性能改善从3.3%提高到4.2%。这充分表明,模糊控制算法相比随机算法有更好的控制效果。The high power density in current multi-core microprocessors results in dramatic thermal fluctuations, which impair the processor＇s reliability and reduces its lifetime. In the paper, we propose a fuzzy control algorithm to make scheduling decisions on task migration. Through simulation analysis and implementation embedded into Linux 2. 6. 8. 1 kernel scheduler on a P4 dual-pore processor, we lower the peak temperature successfully. We further choose random scheduling algorithm as reference to evaluate the efficiency of our fuzzy control algorithm. The experiment results show that the proportion of thermal limit violations decreases from 77 % to 64% and performance improvement increase from 3.3% to 4. 2%. These facts indicate our fuzzy control algorithm outperforms the random scheduling algorithm.

关 键 词：模糊控制 多核处理器 温度感知 动态调度 

分 类 号：TP273.4[自动化与计算机技术—检测技术与自动化装置]