臺灣博碩士論文加值系統

多核心系統在現代科技的應用已越來越普遍，舉凡自小型的手持裝置、筆記型電腦、桌上型電腦甚至於大型資料中心皆已使用多核心系統架構。然而，在多核心系統中執行平行程式可能會因為程式平行化的程度的影響使得每個核心所被分配到的工作是不平均的，進而可能造成有部分的核心沒有被善加利用導致電源的浪費。

在此篇論文我們提出一電源管控方法，用以改善在多核心系統架構上執行OpenMP程式的整體系統之功率消耗。其方法主要是在平行迴圈（Parallel-for loop）中，考慮到多核心上的每個核心所被分配到的工作量的不同，將未善加利用的核心在盡量不影響效能以及計算結果的前提之下，以電源匣控（Power-Gating）技術關閉以降低整體系統之功率消耗。實驗結果顯示我們提出的電源管控方法分別對於Rodinia benchmark suite以及NAS Parallel Benchmarks裡的程式可以有效的減少平均6.5%以及3.9%的耗能。

Multi-core processors are now widely used in modern computing, from embedded to large- scale computing systems. The computing power of such systems is noticeably increased as more cores being integrated in a chip and multiple threads running in parallel. However, the multiple cores might not be utilized efficiently due to the degree of parallelism of work-loads; that is, the workloads distributed to these multiple cores might be imbalanced, which reveals an opportunity for reducing the energy consumption that is dissipated when a core is idle. In this thesis, we present a history-based OpenMP power-control mechanism, called OpenMP-HBPCM, to reduce the energy consumption of OpenMP applications that run on multi-core machines. The mechanism includes two major parts: (1) a compile-time instrumentation that inserts the codes for energy management into possible parallel-for loops and (2) an OpenMP power-control library that decides whether to apply a power control to a core upon a history-based prediction. The experimental results demonstrate that the proposed history-based OpenMP power-control mechanism reduced energy consumption with only little performance degradation. The geometric average reduction in the energy consumption is 6.5% for the Rodinia benchmark suite (five selected benchmarks), and 3.9% for NAS Parallel Benchmarks, while the geometric average of performance degradation is only 0.7% for the Rodinia benchmark suite, and 1.4% for NAS Parallel Benchmarks.

1 Introduction 1
1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Contribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4 Thesis Organization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2 Introduction to OpenMP 6
2.1 OpenMP History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.2 OpenMP Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.3 OpenMP Programming Model . . . . . . . . . . . . . . . . . . . . . . . 12
2.4 OpenMP Loop Iterations Scheduling Policy . . . . . . . . . . . . . . . . 13
3 History-Based OpenMP Power-Control Mechanism 15
3.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.2 Compile-Time Instrumentation for Energy Management . . . . . . . . . . 17
3.3 OpenMP Power-Control Library . . . . . . . . . . . . . . . . . . . . . . 21
3.3.1 History-based Execution Time Prediction . . . . . . . . . . . . . 23
3.3.2 Power Control Determination . . . . . . . . . . . . . . . . . . . 24
4 Experimental Results and Discussion 28
4.1 Experimental Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
4.1.1 Evaluation Platform and Benchmark Suites . . . . . . . . . . . . 28
4.2 Evaluation and Discussions . . . . . . . . . . . . . . . . . . . . . . . . . 30
4.2.1 Normalized Energy Consumption . . . . . . . . . . . . . . . . . 30
4.2.2 Normalized Execution Time . . . . . . . . . . . . . . . . . . . . 33
4.2.3 Prediction Accuracy . . . . . . . . . . . . . . . . . . . . . . . . 35
4.2.4 Energy Delay Product . . . . . . . . . . . . . . . . . . . . . . . 38
5 RelatedWork 41
6 Conclusions and FutureWork 43
6.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
6.2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

[1] A guide to OpenMP. http://www.cs.uh.edu/˜hpctools/OpenMP.
[2] Open64 official website. http://www.open64.net/.
[3] OpenMP wiki. http://en.wikipedia.org/wiki/OpenMP.
[4] Rodinia wiki. http://www.cs.virginia.edu˜skadron/wiki/rodinia/index.php/Main_Page.
[5] Blaise Barney. OpenMP tutorial.
http://computing.llnl.gov/ tutorials/openMP/.
[6] Blaise Barney. Parallel computing tutorial. http://computing.llnl.gov/
tutorials/parallel_comp/.
[7] Shuai Che, M. Boyer, Jiayuan Meng, D. Tarjan, J.W. Sheaffer, Sang-Ha Lee, and K. Skadron. Rodinia: A benchmark suite for heterogeneous computing. In Workload Characterization, 2009. IISWC 2009. IEEE International Symposium on, pages 44–54, oct. 2009.
[8] Juan Chen, Yong Dong, Xuejun Yang, and Panfeng Wang. Energy constrained OpenMP static loop scheduling. In High Performance Computing and Communications, 2008. HPCC ’08. 10th IEEE International Conference on Date of Conference: 25-27 Sept. 2008, pages 139 –146, sept. 2008.
[9] NASA Advanced Supercomputing Division. NAS Parallel Benchmarks. http://www.nas.nasa.gov/publications/npb.html.
[10] James Donald and Margaret Martonosi. Power efficiency for variation-tolerant multicore processors. In Proceedings of the 2006 international symposium on Low power electronics and design, ISLPED ’06, pages 304–309, New York, NY, USA, 2006. ACM.
[11] Jungseob Lee and Nam Sung Kim. Analyzing potential throughput improvement of power- and thermal-constrained multicore processors by exploiting DVFS and PCPG. Very Large Scale Integration (VLSI) Systems, IEEE Transactions on, 20(2):225 –235, feb. 2012.
[12] Matteo Monchiero, Ramon Canal, and Antonio Gonz’alez. Design space exploration for multicore architectures: a power/performance/thermal view. In Proceedings of the
20th annual international conference on Supercomputing, ICS ’06, pages 177–186, New York, NY, USA, 2006. ACM.
[13] Enric Musoll. Energy and thermal tradeoffs in hardware-based load balancing for clustered multi-core architectures implementing power gating. In Proceedings of the 2008 Symposium on Application Specific Processors, SASP ’08, pages 89–94, Washington, DC, USA, 2008. IEEE Computer Society.
[14] OpenMP Architecture Review Board (ARB). OpenMP Application Program Interface, 3.0 edition, May 2008.
[15] Ravishankar Rao, Sarma Vrudhula, and Chaitali Chakrabarti. Throughput of multicore processors under thermal constraints. In Proceedings of the 2007 international symposium on Low power electronics and design, ISLPED ’07, pages 201–206, New York, NY, USA, 2007. ACM.
[16] Lukasz G. Szafaryn, Todd Gamblin, Bronis R. De Supinski, and Kevin Skadron. Experiences with achieving portability across heterogeneous architectures. In Proceedings of the First International Workshop on Domain-Specific Languages and High-Level Frameworks for High Performance Computing (WOLFHPC), in conjunction with ICS, May 2011.