4 \chapter*{Abstract \markboth{Abstract}{Abstract}}
6 \addcontentsline{toc}{chapter}{Abstract}
8 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
12 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
15 \emph{ \begin{center} \Large Energy Consumption Optimization of Parallel Applications with Iterations using CPU Frequency Scaling\end{center}}
16 %\emph{ \begin{center} \large By \end{center}}
17 \emph{ \begin{center} \large Ahmed Badri Muslim FANFAKH \\ University of Franche-Comt\'e, 2016 \end{center}}
18 %\emph{ \begin{center} \large The University of Franche-Comt\'e, 2015 \end{center}}
19 \emph{ \begin{center} \large Supervisors: Raphaël Couturier and Jean-Claude Charr \end{center}}
22 In recent years, green computing has become an important topic in the supercomputing
23 research domain. However, the computing platforms are still consuming more and more energy due to the increase in the number of nodes composing them.
24 To minimize the operating costs of these platforms many techniques have
25 been used. Dynamic voltage and frequency scaling (DVFS) is one of them. It
26 can be used to reduce the power consumption of the CPU while computing, by
27 lowering its frequency. However, lowering the frequency of a CPU may increase
28 the execution time of the application running on that processor. Therefore, the
29 frequency that gives the best trade-off between the energy consumption and the
30 performance of an application must be selected.
31 This thesis, presents the algorithms developed to optimize the energy consumption
32 and the performance of synchronous and asynchronous message passing applications with iterations running over clusters or grids. The energy consumption and performance models for each type of parallel application predicts its execution time and energy consumption for any selected frequency
33 according to the characteristics of both the application and the architecture executing this application.
36 The contribution of this thesis can be divided into three parts: Firstly, optimizing the trade-off between
37 the energy consumption and the performance of the message passing applications with synchronous iterations
38 running over homogeneous clusters. Secondly, adapting the energy and performance models to heterogeneous platforms where each node can have different specifications such as computing power, energy consumption, available frequency gears or network's latency and bandwidth. The frequency scaling algorithm was also modified to suit the heterogeneity of the platform. Thirdly, the models and the frequency scaling algorithm were completely rethought to take into considerations the asynchronism in the communication and computation.
39 All these models and algorithms were applied to message passing applications with iterations and evaluated
40 over either SimGrid simulator or Grid'5000 platform. The experiments showed that the proposed algorithms are
41 efficient and outperform existing methods such as the energy and delay product. They also introduce a small
42 runtime overhead and work online without any training or profiling.
44 \textbf{KEY WORDS:} Dynamic voltage and frequency scaling, Grid computing, Energy optimization, parallel applications with iterations and online frequency scaling algorithm.