以執行緒為觀點之預測模型

A Thread View Prediction Model

蔡宗祐(Tsung-Yu Tsai)

指導教授 ： 黃婷婷

國立清華大學/電機資訊學院/資訊工程學系/碩士(2012年)

摘要

為了有效利用多處理器晶片(chip multiprocessor, CMPs)，吾人以多元的技術發展平行化工作負荷(parallel workloads)，管線化之平行模型即為其中一種平行程序化的樣式。為了對於上述程序編制的模式近一步地研究平行架構與平行程序最佳化，瞭解不同多執行緒應用程序(multi-threaded application)的行為將顯得重要。本論文對於thread-based model、collective-thread-based model、collective-core-based model以及core-based model四種不同觀點之行為預測模型做出評價，前二者以執行緒之觀點分析應用程序的行為並做出預測，後兩者則以硬體的觀點分析行為加以預測。此評量結果顯示出隨著平行化工作負荷的發展，以執行緒為觀點之預測模型顯得更準確可靠與重要。

關鍵字

執行緒 ； 預測

並列摘要

Parallel workloads are developed with diverse techniques and methods to take advantage of chip multiprocessor (CMPs). The pipeline parallelization model is one such method that is an increasingly popular parallel programming pattern for emerging applications. With this trend of programming fashion, understanding the behavior of various multi-threaded applications is important for research topics of parallel architecture and parallel program optimization. This thesis gives an evaluation of four different views of behavior analysis, the thread-based model, the collective-thread-based model, the collective-core-based model and the core-based model. The thread-based model and the collective-thread-based model collect the application behavior with the thread perspective. The collective-core-based model and the core-based model collect the application behavior with the core perspective. The evaluation shows that the models with the thread perspective has advantage as compared to the models with the core perspective, which demonstrates that the analysis of application behavior with the thread perspective will be more important and faithful with the development of parallel workloads.

並列關鍵字

Thread ； Prediction

參考文獻

[2] C. Bienia and K. Li, “Characteristics of workloads using the pipeline programming model,” in Proceedings of the 3rd Workshop on Emerging Applications and Many-core Architecture, 2010.

[3] Y. Kim, D. Han, O. Mutlu, and M. Harchol-Balter, “Atlas: A scalable and high-performance scheduling algorithm for multiple memory controllers,” in High Performance Computer Architecture (HPCA), 2010 IEEE 16th International Symposium on, 2010, pp. 1–12.

[6] C. Isci, G. Contreras, and M. Martonosi, “Live, runtime phase monitoring and prediction on real systems with application to dynamic power management,” in Microarchitecture, 2006. MICRO-39. 39th Annual IEEE/ACM International Symposium on, 2006, pp. 359 –370.

[7] E. Duesterwald, C. Cascaval, and S. Dwarkadas, “Characterizing and predicting program behavior and its variability,” in Parallel Architectures and Compilation Techniques, 2003. PACT 2003. Proceedings. 12th International Conference on, 2003, pp. 220 – 231.

[8] R. Sarikaya and A. Buyuktosunoglu, “A unified prediction method for predicting program behavior,” Computers, IEEE Transactions on, pp.272 –282, 2010.