資料中心之伺服器內部配置多顆風扇與發熱元件，不同風扇轉速、發熱元件佈局與負載高低皆影響散熱流場，故伺服器散熱控制被歸類為多輸入多輸出非線性問題，不易以簡單線性系統的模式進行散熱分析與控制。本研究致力於建立一套用於伺服器散熱的多風扇控制系統，以權重機制之非同步調變多顆風扇，結合演化策略演算法並設計適應度函式，實現不同權重之多風扇控制系統。所研發的控制系統可根據不同伺服器負載，在有限搜索數內取得近似最佳權重選項，進行非同步風扇轉速調變，在符合伺服器最高溫度設定條件下，節能近似最佳化。本研究所提出的散熱控制系統，設計八種不同的伺服器負載情境之實驗，相較於同步風扇轉速調變，平均節省43.1%的風扇總功率，而且在所有權重中僅需實測14.2%的權重選項。另外，各個近似最佳權重屬於對應的全域最佳解發生機率達47.5%。實驗結果顯示，本研究所提出的不同權重之多風扇控制系統，能同時滿足伺服器最高溫度限制與節能近似最佳化，系統效能優異且無須建構熱傳模型，亦不需蒐集大量資料集，即可進行散熱分析。因此，建構系統成本低，未來更改內部程式碼參數，即可用於不同種類伺服器之散熱問題。

Each server in data centers is equipped with multiple fans and heatgenerating components. Factors such as fans speed, layout of components, and loading scenarios, will influence the airflow and thermal flow fields within a server. Hence, a server thermal control is classified as a MultiInput Multi-Output (MIMO) nonlinear control system, which cannot be analyzed and controlled using simple linear system method. By using the weighting mechanism of different weighting values for each cooling fan to asynchronously modulate multiple fans, and combining the Evolutionary Strategy (ES) algorithm with designing fitness functions, this study realizes the multi-fan thermal control system for a server. The developed system can attain approximately optimal weightings within a limited number of searches, based on different loading scenarios. By modulating fans asynchronously with these weightings, the system can achieve approximately optimal energy-saving while still meet the thermal specifications in a server; that is the allowed highest temperature of CPU and PCIe. Compared to modulating fans synchronously, the multi-fan control system modulating fans asynchronously saves an average of 43.1% of the total fans power, and only need to search 14.2% of all weighting options under the eight designed loading scenarios for experiments. Furthermore, the probability of each approximate optimal weighting corresponding to global optimal solution is 47.5%. Experimental results demonstrate that the proposed asynchronously modulating multi-fan control system can simultaneously satisfy the thermal specifications and achieves approximately optimal energy-saving for a server. As a result, the developed system is feasible with excellent performance for significant energy saving, while it is no need to construct a mathematical thermal models or to analyze numerous datasets. In the future, by adjusting the code parameters of system, it may be to be applied to various types of servers.

摘要 i
Abstract ii
誌謝 iii
圖目錄 vi
表目錄 ix
第一章 緒論 1
1 1.1 前言 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 研究動機及目的 . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 文獻回顧 . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.4 本文架構 . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
第二章 伺服器與相關設備 11
2.1 伺服器規格 . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 內建軟體 . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2.1 PTU . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2.2 智慧型平台管理介面 . . . . . . . . . . . . . . . . 14
2.3 散熱風扇 . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.4 外接設備 . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.4.1 熱電偶 . . . . . . . . . . . . . . . . . . . . . . . . 17
2.4.2 資料擷取器 . . . . . . . . . . . . . . . . . . . . . 17
2.4.3 PWM 訊號產生器 . . . . . . . . . . . . . . . . . . 18
2.4.4 Arduino Mega 2560 與繼電器模組 . . . . . . . . . 19
第三章 伺服器多風扇控制系統實現 21
3.1 權重機制 . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.2 演化策略演算法原理暨應用 . . . . . . . . . . . . . . . . 23
3.3 實驗設計 . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.3.1 伺服器實驗規範 . . . . . . . . . . . . . . . . . . . 32
3.3.2 權重測試實驗 . . . . . . . . . . . . . . . . . . . . 35
3.3.3 權重搜索實驗 . . . . . . . . . . . . . . . . . . . . 35
第四章 實驗結果與討論 37
4.1 權重測試實驗結果與討論 . . . . . . . . . . . . . . . . . . 37
4.2 關鍵發熱元件之熱現象分析 . . . . . . . . . . . . . . . . 52
4.3 權重搜索實驗結果與討論 . . . . . . . . . . . . . . . . . . 61
第五章 結論與未來工作 79
5.1 結論 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
5.2 未來工作 . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
參考文獻 83
附錄 A 權重細調之實測結果 87
附錄 B 權重搜索實驗紀錄(三十回) 101

[1] S. R. Group. (2022) 雲 端 服 務 商 全 球 市 場 收 益 趨 勢.[Online]. Available: https://www.srgresearch.com/articles/q2-cloud-market grows-by-29-despite-strong-currency-headwinds-amazon-increases its-share, accessed: 2022-10-01.
[2] S. R. Group. (2022) 全球大型資料中心數量成長預測.[Online]. Available: https://www.srgresearch.com/articles/pipeline-of-over 300-new-hyperscale-data-centers-drives-healthy-growth-forecasts, accessed: 2022-10-02.
[3] R. Hintemann and S. Hinterholzer, “Energy consumption of data centers worldwide-how will the internet become green?” ICT4S” Lappeenranta, Finnland, June 10–14, 2019.
[4] I. E. Agency. (2022) 2016 年度各國用電量. [Online]. Available: https://www.iea.org/data-and-statistics, accessed: 2022-10-03.
[5] X. Wei, J. Xiao, Z. Wang, Z. Wang, and Y. Tian, “Integration and development of energy and information network in the pan-arctic region,” Global Energy Interconnection, vol. 2, no. 6, pp. 504–512, 2019.
[6] AKCP. (2022) 資料中心電力使用量. [Online]. Available: https: //www.akcp.com/blog/the-real-amount-of-energy-a-data-center-use/, accessed: 2022-10-04.
[7] P. Agreement, “Paris agreement,” Report of the Conference of the Parties to the United Nations Framework Convention on Climate Change (21st Session, 2015: Paris). Retrived December, vol. 4, p. 2017, Paris, France, November 30 - December 13, 2015.
[8] M. Lennan and E. Morgera, “The glasgow climate conference (cop26),” The International Journal of Marine and Coastal Law, vol. 37, no. 1, pp. 137–151, 2022.
[9] T. Grid, “Green grid metrics: Describing datacenter power efficiency,” Online at http://www. thegreengrid. org/gg_content/Green_Grid_Metrics_WP. pdf, 2007.
[10] Q. Zhang, Z. Meng, X. Hong, Y. Zhan, J. Liu, J. Dong, T. Bai, J. Niu, and M. J. Deen, “A survey on data center cooling systems: Technology, power consumption modeling and control strategy optimization,” Journal of Systems Architecture, vol. 119, p. 102253, 2021.
[11] Z. Wang, C. Bash, N. Tolia, M. Marwah, X. Zhu, and P. Ranganathan, “Optimal fan speed control for thermal management of servers,” International Electronic Packaging Technical Conference and Exhibition, vol. 43604, pp. 709–719, San Francisco, California, USA, July 19 - 23, 2009.
[12] D. Shin, J. Kim, N. Chang, J. Choi, S. W. Chung, and E.-Y. Chung, “Energy-optimal dynamic thermal management for green computing,” 2009 IEEE/ACM International Conference on Computer Aided Design-Digest of Technical Papers, pp. 652–657, San Jose, CA, USA, November 2 - 5, 2009.
[13] C. Lee and R. Chen, “Optimal self-tuning pid controller based on low power consumption for a server fan cooling system,” Sensors, vol. 15, no. 5, pp. 11 685–11 700, 2015.
[14] B. Acun, “Mitigating variability in hpc systems and applications for performance and power efficiency,” Ph.D. dissertation, University of Illinois at Urbana-Champaign, 2017.
[15] 林炘泓，“GRU 實現伺服器 PCIE 溫度估測器及散熱控制系統＂， 國立清華大學碩士論文，2020 年 7 月。
[16] W.-X. Chu, Y.-H. Lien, K.-R. Huang, and C.-C. Wang, “Energy saving of fans in air-cooled server via deep reinforcement learning algorithm,” Energy Reports, vol. 7, pp. 3437–3448, 2021.
[17] T. A. Byquist, and M. T. Crocker,“Method and system of advance fan speed control,＂U.S Patent, US 7.490,479 B2, 2009.
[18] 朱信承，“應用深度學習演算法於伺服器 PCIE 之溫度解耦合控 制系統＂，國立清華大學碩士論文，2021 年 8 月。
[19] EIA. (2007) Define: Eia-310. [Online]. Available: https://www. server-racks.com/eia-310.html, accessed: 2022-10-05.
[20] Intel. (2014) Platform power thermal utility. [Online]. Available: https://designintools.intel.com/ platform-power-thermal-utility-ptu-aka-maximum-power-program. html, accessed: 2022-10-06.
[21] Intel. (2013) Intelligent platform management interface specification. [Online]. Available: https: //www.intel.com/content/www/us/en/products/docs/servers/ipmi/ipmi-second-gen-interface-spec-v2-rev1-1.html, accessed: 2022-10- 07.
[22] Delta. (2015) Pfb series. [Online]. Available: https: //filecenter.deltaww.com/Products/Download/03/0301/Catalogue/ PFR80x80x38mm.pdf, accessed: 2022-10-08.
[23] M. Paulsson and S. Datta, “Thermoelectric effect in molecular electronics,” Physical Review B, vol. 67, no. 24, p. 241403, 2003.
[24] Keysight. (2010) 34970a data acquisition/data logger switch unit. [Online]. Available: https://www.keysight.com/us/en/product/ 34970A/34970a-data-acquisition-control-mainframe-modules.html, accessed: 2022-10-09.
[25] Arduino. (2021) Mega 2560 rev3. [Online]. Available: https: //store.arduino.cc/products/arduino-mega-2560-rev3, accessed: 2022- 10-10.
[26] K. H. Ang, G. Chong, and Y. Li, “Pid control system analysis, design, and technology,” IEEE transactions on control systems technology, vol. 13, no. 4, pp. 559–576, 2005.
[27] A. E. Eiben, J. E. Smith et al., Introduction to evolutionary computing. Springer, 2003, vol. 53.
[28] C. Darwin, On the origin of species, 1859. Routledge, 2004.
[29] H.-G. Beyer and H.-P. Schwefel, “Evolution strategies–a comprehensive introduction,” Natural computing, vol. 1, pp. 3– 52, 2002.
[30] S. Kern, S. D. Müller, N. Hansen, D. Büche, J. Ocenasek, and P. Koumoutsakos, “Learning probability distributions in continuous evolutionary algorithms–a comparative review,” Natural Computing, vol. 3, no. 1, pp. 77–112, 2004.
[31] H.-P. Schwefel, Numerical optimization of computer models. John Wiley & Sons, Inc., 1981. [32] N. Hansen and A. Ostermeier, “Completely derandomized self adaptation in evolution strategies,” Evolutionary computation, vol. 9, no. 2, pp. 159–195, 2001.
[33] N. Hansen, “The cma evolution strategy: A tutorial,” arXiv preprint arXiv:1604.00772, 2016.
[34] M. Lunacek, D. Whitley, and A. Sutton, “The impact of global structure on search,” Parallel Problem Solving from Nature–PPSN X: 10th International Conference, Proceedings 10, pp. 498–507, Dortmund, Germany, September 13 - 17, 2008.
[35] S. R. Gagganapalli, “Implementation and evaluation of cma-es algorithm,” Master’s thesis, North Dakota State University, 2015.
[36] T. Akiba, S. Sano, T. Yanase, T. Ohta, and M. Koyama, “Optuna: A next-generation hyperparameter optimization framework,” Proceedings of the 25th ACM SIGKDD international conference on knowledge discovery & data mining, pp. 2623–2631, Anchorage, AK, USA, August 4 - 8, 2019.
[37] L. R. Gay and P. Diehl, Research methods for business and management. Macmillan Coll Division, 1992.
[38] W.-X. Chu, R. Wang, P.-H. Hsu, and C.-C. Wang, “Assessment on rack intake flowrate uniformity of data center with cold aisle containment configuration,” Journal of Building Engineering, vol. 30, p. 101331, 2020. [39] V. A. Epanechnikov, “Non-parametric estimation of a multivariate probability density,” Theory of Probability & Its Applications, vol. 14, no. 1, pp. 153–158, 1969.