王勝德臺灣大學：電機工程學研究所陳宏明Chen, Hung-MingHung-MingChen2007-11-262018-07-062007-11-262018-07-062004http://ntur.lib.ntu.edu.tw//handle/246246/53525在這篇論文中，我們提出非貪婪(non-greedy)的處理器動態電壓調整(Dynamic Voltage Scaling)演算法，稱為”整合處理器與周邊裝置動態電壓調整” (Integrated processor and devices DVS, IDVS)。它針對減少過去動態電壓調節技術儘可能貪婪地調降處理器速度所引起額外的周邊裝置耗能。在過去的研究中，僅僅考慮處理器減速來降低處理器耗能，卻忽略了工作的執行可能會使用到其他的周邊裝置，處理器減速造成工作執行時間變長，使得週邊裝置的使用時間增加。這將有可能會造成在處理器上省下的能源反而從週邊裝置浪費掉。為了做整個系統處理器與裝置的排程，我們分別修改了MUSCLES與LEDES兩種週邊裝置排程的演算法來與我們的處理器動態電壓調整演算法一起工作，藉以觀察延長工作時間對於整個系統耗能的影響。In this thesis, we present a class of non-greedy dynamic voltage scaling algorithms called Integrated processor and devices DVS (IDVS) that is aimed to reduce extra energy consumption caused by the greedy DVS approach. The greedy DVS method tries to reduce the processor power consumption but may increase devices working time. The greedy DVS algorithm simply exploits available slack time while the proposed non-greedy DVS determines the processor operating speed and voltage with the consideration of the used I/O device power consumption. Therefore, it tried to determine the optimal task execution time that minimizes energy consumption of the overall system. We also modified the device scheduling algorithms MUSCLES and LEDES and we made them work with DVS technique to save energy consumption of processor and devices. The proposed approach is called look ahead LEDES (L-LEDES) and look ahead MUSCLES (L-MUSCLES), and they guarantee that no task deadlines are missed due to device transition overhead. L-MUSCLES makes use of a technique that gathers long enough pieces of task execution time or inter-task time and determines the devices wake-up/sleeping time at legal scheduling instant. Simulation results show that the proposed algorithms can reduce energy consumption of the overall system for real-time systems without violating real-time requirements.中文摘要 i 英文摘要 iii 謝辭 v 目錄 vii 表列 ix 圖列 xi 1. Introduction 1 2. Related Works 3 2.1 Dynamic voltage scaling 3 2.2 Dynamic power management for I/O device 4 3. Model and parameter 5 3.1 system model 5 3.2 parameter and definition 5 4. Motivational example and problem 7 4.1 Problem with greedy DVS 9 4.2 Problem with devices scheduling 12 5. Reduce energy consumption of overall system 14 5.1 DVS and non-DVS section 15 6. Integrated processor and devices DVS 16 6.1 Non-greedy DVS and greedy DVS 16 6.2 Slack Reclamation 18 6.3 Overhead conscious with adjust speed and voltage 18 6.3.1 Time overhead 18 6.3.2 Energy overhead 20 6.4 Advance energy saving in processor 21 6.5 Energy-efficient scaling 22 7. L-MUSCLES devices scheduling 24 7.1 Review MUSCLES 24 7.2 L-MUSCLES mechanism 27 7.2.1 History-based learning 27 7.2.2 Piece collection 28 7.2.3 Floating scheduling instant 29 8. Experiments 32 8.1 Methodology 32 8.2 Pattern and parameters 33 8.3 Test plan 35 8.4 Task-based power management procedure and architecture 36 8.5 Simulation 37 9. Conclusions 45 10. Reference 47607044 bytesapplication/pdfen-US即時系統能量耗損電源管理週邊裝置transitionpower managementDVSDPMreal time[SDGs]SDG7thesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/53525/1/ntu-93-R91921017-1.pdf