伺服器的匯流排配接卡(PCIE)常因配線問題而無法安裝溫度感測器，導致PCIE經常發生過熱，影響伺服器的運作。為解決此問題，本研究使用Gate Recurrent Unit(GRU)實現PCIE的溫度估測器，並依據伺服器當前的使用狀態即時估測PCIE溫度，避免其產生過熱。本研究同時研發一散熱控制系統，可分別實現CPU及PCIE的溫度控制，CPU溫度控制為利用試誤法調整散熱風扇的PID控制器參數。PCIE溫度控制包含溫度預測模型以及優化器，溫度預測模型用於預測各風扇轉速下PCIE的未來溫度，利用優化器調整最佳風扇轉速，使PCIE溫度保持在與設定值的最小誤差之下。本研究中，分別以熱電偶的量測以及PCIE估測器取得PCIE溫度，並以此溫度進行散熱控制實驗。於兩種實驗中，二個PCIE卡估測溫度的均方根誤差分別為0.85/0.65 ^o C以及0.69/0.75 ^o C，顯示本研究的PCIE溫度估測器之高性能。而散熱控制系統能夠將CPU及PCIE溫度的溫度超越量保持低於3%，並於發生超越量後分別收斂至80 ^o C及70 ^o C內，確保元件不會因為過熱而影響伺服器的效能。

In a server, the Peripheral Component Interconnect Express (PCIE) is often overheating during operation since no temperature sensor is installed in it due to wiring problems. In this thesis, it implements the PCIE temperature estimator, using the Gate Recurrent Unit (GRU), which can estimate the PCIE temperature in time based on the current server status, to avoid overheating. Meanwhile, the heat dissipation control system is realized to control the temperature of the CPU and PCIE. The CPU temperature control is accomplished by adjusting the PID parameters of fan controller using trial and error in LabVIEW, and the PCIE temperature control includes temperature prediction model and optimizer. The temperature prediction model is used to predict the future temperature of the PCIE at various fan speeds, and the optimizer is used to keep the PCIE temperature at its setting value with minimum error and the optimal fan speed. In this study, the temperatures measuring by thermocouple and estimated by PCIE estimator are used as the inputs to conduct heat dissipation control experiments, respectively. In these two experiments, the root-mean-square error of the estimated temperatures for both PCIEs (PCIEs 1 and 2) are 0.85/0.65 ^o C and 0.69/0.75 ^o C, respectively. Therefore, it indicates the high performance of the developed PCIE temperature estimator. The heat dissipation control system can keep both temperature overshoot of the CPU and PCIE below 3%, and converge to within 80 ^o C and 70 ^o C, respectively, after the overshooting. As a result, it ensures that the key components will not be overheating to further cause the fail in a server.

摘要------------------------------------------I
Abstract-------------------------------------II
圖目錄---------------------------------------VI
表目錄---------------------------------------IX
第一章 緒論-----------------------------------1
1.1前言---------------------------------------1
1.2研究動機-----------------------------------3
1.3文獻回顧-----------------------------------4
1.4本文架構----------------------------------12
第二章 系統架構-------------------------------13
2.1伺服器系統---------------------------------13
2.2系統控制流程-------------------------------15
2.3實驗設備及初始化設定------------------------17
第三章 溫度估測器及散熱控制設計----------------30
3.1特徵資料蒐集-------------------------------30
3.2晶片溫度預測模型設計-----------------------35
3.3散熱控制設計-------------------------------42
第四章 實驗結果-------------------------------46
4.1散熱控制器參數設計-------------------------46
4.2 PCIE溫度估測器之參數設計------------------59
4.3以PCIE實際溫度進行散熱控制之結果討論--------65
4.4以PCIE估測溫度進行散熱控制之結果討論--------74
第五章 結論與未來工作-------------------------84
5.1結論-------------------------------------84
5.2未來工作----------------------------------85
參考文獻-------------------------------------86

[1] State of the IoT 2018: Number of IoT devices now at 7B – Market accelerating. (2018). https://www.iot-analytics.com/state-of-the-iot-update-q1-q2-2018-number-of-iot-devices-now-7b/ (November 1, 2019)
[2] 2020年全球資料量將成長至44ZB(2014)。檢自 https://www.ithome.com.tw/article/87190 (November 1, 2019)
[3] H. Rong, H. Zhang, S. Xiao, C. Li, and C. Hu, "Optimizing energy consumption for data centers", Renewable and Sustainable Energy Reviews, vol. 58, pp. 674-691, January 2016.
[4] J. Gao, "Machine learning applications for data center optimization", Google, White Paper, 2014.
[5] K. Zhang, S. Ogrenci-Memik, G. Memik, K. Yoshii, R. Sankaran, and P. Beckman, "Minimizing Thermal Variation Across System Components", IEEE 29th International Parallel and Distributed Processing Symposium, pp. 1139-1148, May 2015.
[6] K. Zhang, A. Guliani, S. Ogrenci-Memik, G. Memik, K. Yoshii, R. Sankaran, P. Beckman, "Machine Learning-Based Temperature Prediction for Runtime Thermal Management Across System Components", IEEE Transactions on Parallel and Distributed Systems, vol. 29, pp. 405-419, February 2018.
[7] J. Athavale, Y. Joshi, and M. Yoda, "Artificial Neural Network Based Prediction of Temperature and Flow Profile in Data Centers", 17th IEEE Intersociety Conference on Thermal and Thermo-mechanical Phenomena in Electronic Systems (ITherm), pp. 871-880, June 2018.
[8] S. Suzuki, M. Hiraoka, T. Shiraishi, E. Kreshpa, T. Yamamoto, H. Fukuda, S. Matsui, M. Fujisaki, A. Uno, "Power Prediction for High-performance Computing", presented at SC'19, Denver, USA, November 2019.
[9] J. Chen, R. Tan, Y. Wang, G. Xing, X. Wang, X. Wang, B. Punch, D. Colbry, "A High-Fidelity Temperature Distribution Forecasting System for Data Centers", IEEE 33rd Real-Time Systems Symposium, pp. 215-224, December 2012.
[10] J. Chen, R. Tan, G. Xing, and X. Wang, "PTEC: A System for Predictive Thermal and Energy Control in Data Centers", IEEE Real-Time Systems Symposium, pp. 218-227, January 2014.
[11] L. Yao and J.-H. Huang, "Multi-Objective Optimization of Energy Saving Control for Air Conditioning System in Data Center", Energies, vol. 12, p. 1474, April 2019.
[12] T. Shiraishi, H. Endo, T. Hineno, and H. Fukuda, "Robust Data-Driven Power Simulator for Fast Cooling Control Optimization of a Large-Scale Computing System", presented at SC'19, Denver, USA, November 2019.
[13] Z. Wang, C. Bash, N. Tolia, M. Marwah, X. Zhu, P. Ranganathan, "Optimal Fan Speed Control for Thermal Management of Servers", presented at InterPACK'09, San Francisco, USA, July 2009.
[14] J. Kim, M. M. Sabry, D. Atienza, K. Vaidyanathan, and K. Gross, "Global fan speed control considering non-ideal temperature measurements in enterprise servers", Design, Automation & Test in Europe Conference & Exhibition (DATE), pp. 1-6, February 2014.
[15] R. Lucchese, J. Olsson, A.-L. Ljung, W. Garcia-Gabin, and D. Varagnolo, "Energy savings in data centers: A framework for modelling and control of servers’ cooling", IFAC PapersOnLine, vol. 50, pp. 9050-9057, July 2017.
[16] Q. Zheng and Z. Ping, "Active Disturbance Rejection Control for Server Thermal Management", 5th International Conference on Control, Decision and Information Technologies (CoDIT), pp. 158-163, 2018.
[17] Intel® Xeon® 處理器 E5-2699 v3。檢自https://www.intel.com/content/www/tw/zh/ark/products/81061/intel-xeon-processor-e5-2699-v3-45m-cache-2-30-ghz.html (November 1, 2019)
[18] 匯流排系統-MBA智庫百科。檢自https://wiki.mbalib.com/wiki/总线系统 (November 1, 2019)
[19] 34970A 資料擷取/資料記錄切換裝置。檢自https://www.keysight.com/main/techSupport.jspx?pid=1000001313:epsg:pro&t=79841.g.1&pageMode=PL&lc=cht&cc=TW (November 1, 2019)
[20] LabVIEW2019. https:// www.ni.com/ (November 1, 2019)
[21] Python. https://www.python.org/ (November 1, 2019)
[22] https://www.labviewmakerhub.com/doku.php?id=libraries:linx:start (November 1, 2019)
[23] ZeroMQ. https://www.zeromq.org/ (April 16, 2020)
[24] Pyrorch. https://www.pytorch.org/ (January 25, 2020)
[25] R. Pascanu, T. Mikolov, and Y. Bengio, "On the difficulty of training recurrent neural networks", 30th International Conference on Machine Learning, February 2013.
[26] S. Hochreiter and J. Schmidhuber, "Long Short-term Memory", Neural Computation, vol. 9, pp. 1735-80, December 1997.
[27] K. Cho, B. van Merrienboer, D. Bahdanau, Y. Bengio, "On the Properties of Neural Machine Translation: Encoder-Decoder Approaches", Eighth Workshop on Syntax, Semantics and Structure in Statistical Translation(SSST-8), October 2014.