隨著科技的進步，個人裝置（例如筆記型電腦與智慧型手機）擁有較以往更佳的硬體效能。同時，各種不同的多媒體應用對於運算資源產生了逐漸增加的需求。對於這樣的情境，我們採用多媒體霧計算（fog computing）平臺的概念，目標為減少使用雲端計算平臺的成本。在此平臺中，服務提供者（fog provider）接收來自於服務使用者（fog users）的工作，並分配給工作者／裝置（fog workers/devices）。對於實現此概念，主要有三個研究議題： （一）預測完成工作所需的資源、（二）預測裝置可提供的資源、（三）對於工作與可用資源進行排程。本論文主要針對預測裝置可提供的資源進行研究。我們採用三個機器學習演算法：隨機森林（Random Forest）、梯度提升樹（Gradient Boosting Tree）與神經網路（Neural Network），並使用開源函式庫實作。我們使用兩組資料：使用者資料（desktop dataset）與數據中心資料（datacenter dataset），兩者的資源紀錄分別來自於真實使用者與數據中心裡的機器。我們使用兩組資料的4/5進行10次交叉驗證，得出機器學習演算法所需的超參數（hyperparameter）。結果顯示兩組資料所需的最佳超參數是不同的。從此可知，當服務提供者採用新的資料，或是資料有大量變異時，必須重新微調超參數。我們使用兩組資料剩下的1/5以及真實的動畫處理資料來進行模擬。實驗結果顯示：（一）神經網路演算法對於兩組資料可達到預測值與實際值分別僅6.08%與2.00%的差異、（二）較準確的可用資源量預測可使失敗的工作量減少。

The personal devices such as laptops and smartphones are being equipped with better hardware, which leads to stronger computing abilities. At the same time, the demand of variousmultimedia applications requires increasing computational resources. We propose to build the multimedia fog computing platform, which aims at reducing the cost of using cloud computing. In this platform, the fog provider receives the jobs from the fog users and schedules them to the fog workers/devices. There are three main research problems: (i) prediction of the required amount of resources of the jobs, (ii) prediction of the available resources of the fog devices, and (iii) scheduling the jobs and the fog devices. This thesis focuses on the prediction of the amount of the available resources. We adopt three machine learning algorithms, namely, the Random Forest, Gradient Boosting Tree, and Neural Network, and implement them using open source libraries. We apply two datasets, desktop and datacenter datasets, where the traces come from real users and machines in
the cloud datacenter, respectively. We use 80% of both datasets and perform 10-fold cross validation to fine-tune the hyperparameters of the proposed algorithms. The optimal combinations of the hyperparameters for both datasets are different. We learned that when the fog provider applies new datasets, or when the dataset dramatically changes, it is necessary to re-tune the hyperparameters. We implement a simulator and use the rest 20% of both available resource datasets and a real animation rendering jobs dataset to drive our simulator. The simulation results show that: (i) the Neural Network-based algorithm achieves 6.08% and 2.00% deviation in average for the desktop and datacenter datasets, respectively, and (ii) more accurate prediction of the amount of available resources leads to fewer failed jobs.

[1] BitBrains. https://www.bitbrains.nl/solvinity-en.
[2] Disney rendered its new animated film on a 55,000-core supercomputer. https://www.engadget.com/2014/10/18/disney-big-hero-6/.
[3] Microsoft developer network (MSDN). https://msdn.microsoft.com/en-us/.
[4] Pixar by the numbers - from toy story to brave. http://collider.com/pixar-numbers-toy-story-brave/.
[5] Pixar by the numbers - from toy story to monsters university. http://collider.com/pixar-numbers-monsters-university/.
[6] Scikit-learn. http://scikit-learn.org.
[7] SETI@home. http://setiathome.berkeley.edu/.
[8] XGBoost. https://github.com/dmlc/xgboost.
[9] Y. S. Abu-Mostafa, M. Magdon-Ismail, and H. Lin. Learning from data, volume 4. AMLBook Singapore, 2012.
[10] S. Akioka and Y. Muraoka. Extended forecast of cpu and network load on computational grid. In Cluster Computing and the Grid, 2004. CCGrid 2004. IEEE International Symposium on, pages 765–772. IEEE, 2004.
[11] D. P. Anderson. BOINC: A system for public-resource computing and storage. In Grid Computing, 2004. Proceedings. Fifth IEEE/ACM International Workshop on, pages 4–10. IEEE, 2004.
[12] D. P. Anderson, J. Cobb, E. Korpela, M. Lebofsky, and D. Werthimer. SETI@home: an experiment in public-resource computing. Communications of the ACM, 45(11):56–61, 2002. 32
[13] A. Andrzejak, D. Kondo, and D. P. Anderson. Ensuring collective availability in volatile resource pools via forecasting. In Managing Large-Scale Service Deployment, pages 149–161. Springer, 2008.
[14] F. Bonomi, R. Milito, J. Zhu, and S. Addepalli. Fog computing and its role in the internet of things. In Proc. of ACM Workshop on Mobile Cloud Computing (MCC), pages 13–16, Helsinki, Finland, August 2012.
[15] J. Brevik, D. Nurmi, and R. Wolski. Automatic methods for predicting machine availability in desktop grid and peer-to-peer systems. In Cluster Computing and the Grid, 2004. CCGrid 2004. IEEE International Symposium on, pages 190–199. IEEE, 2004.
[16] P. Brucker and S. Knust. Complex Scheduling. Springer, 2012.
[17] M. Carvalho, R. Miceli, P. D. M. Jr, F. Brasileiro, and R. Lopes. Predicting the quality of service of a peer-to-peer desktop grid. In Proceedings of the 2010 10th IEEE/ACM International Conference on Cluster, Cloud and Grid Computing, pages 649–654. IEEE Computer Society, 2010.
[18] C. Dabrowski and F. Hunt. Using markov chain analysis to study dynamic behaviour in large-scale grid systems. In Proceedings of the Seventh Australasian Symposium on Grid Computing and e-Research-Volume 99, pages 29–40. Australian Computer Society Inc., 2009.
[19] S. Di, D. Kondo, andW. Cirne. Google hostload prediction based on Bayesianmodel with optimized feature combination. Journal of Parallel and Distributed Computing, 74(1):1820–1832, 2014.
[20] N. D. Doulamis, A. D. Doulamis, A. Panagakis, K. Dolkas, T. A. Varvarigou, and E. Varvarigos. A combined fuzzy-neural network model for non-linear prediction of 3-D rendering workload in grid computing. IEEE Transactions on Systems, Man, and Cybernetics, Part B (Cybernetics), 34(2):1235–1247, 2004.
[21] J. Friedman, R. Tibshirani, and T. Hastie. The Elements of Statistical Learning: Data Mining, Inference, and Prediction. Springer, 2001.
[22] D. Gmach, J. Rolia, L. Cherkasova, and A. Kemper. Workload analysis and demand prediction of enterprise data center applications. In 2007 IEEE 10th International Symposium on Workload Characterization, pages 171–180. IEEE, 2007. 33
[23] H. Hong, J. Chuang, and C. Hsu. Animation rendering on multimedia fog computing platforms. In IEEE 8th International Conference on Cloud Computing Technology and Science (CloudCom). IEEE, 2016.
[24] B. Javadi, D. Kondo, J. Vincent, and D. P. Anderson. Discovering statistical models of availability in large distributed systems: An empirical study of seti@ home. Parallel and Distributed Systems, IEEE Transactions on, 22(11):1896–1903, 2011.
[25] R. E. Kalman. A new approach to linear filtering and prediction problems. Journal of basic Engineering, 82(1):35–45, 1960.
[26] D. Kondo, A. Andrzejak, and D. P. Anderson. On correlated availability in internet distributed systems. In Proceedings of the 2008 9th IEEE/ACM International Conference on Grid Computing, pages 276–283. IEEE Computer Society, 2008.
[27] S. J. Orfanidis. Introduction to Signal Processing. Prentice-Hall, Inc., 1995.
[28] K. Ramachandran, H. Lutfiyya, and M. Perry. Decentralized approach to resource availability prediction using group availability in a p2p desktop grid. Future Generation Computer Systems, 28(6):854–860, 2012.
[29] B. Rood and M. J. Lewis. Grid resource availability prediction-based scheduling and task replication. Journal of Grid Computing, 7(4):479–500, 2009.
[30] L. Sarmenta. Volunteer computing. PhD thesis, Massachusetts Institute of Technology, 2001.
[31] L. Shang, Z.Wang, X. Zhou, X. Huang, and Y. Cheng. TM-DG: a trust model based on computer users’ daily behavior for desktop grid platform. In Proceedings of the 2007 symposium on Component and framework technology in high-performance and scientific computing, pages 59–66. ACM, 2007.
[32] L. Vaquero and L. Merino. Finding your way in the fog: Towards a comprehensive definition of fog computing. ACM SIGCOMM Computer Communication Review, 44(5):27–32, October 2014.
[33] Y. Wu, Y. Yuan, G. Yang, and W. Zheng. Load prediction using hybrid model for computational grid. In 2007 8th IEEE/ACM International Conference on Grid Computing, pages 235–242. IEEE, 2007.
[34] D. Yang, J. Cao, C. Yu, and J. Xiao. A multi-step-ahead CPU load prediction approach in distributed system. In Cloud and Green Computing (CGC), 2012 Second International Conference on, pages 206–213. IEEE, 2012. 34
[35] Y. Yuan, Y. Wu, G. Yang, and W. Zheng. Adaptive hybrid model for long term load prediction in computational grid. In Cluster Computing and the Grid, 2008. CCGRID’08. 8th IEEE International Symposium on, pages 340–347. IEEE, 2008.