Scientific workflows decompose complex scientific applications into smaller subsequent interdependent tasks that can be executed in serial or parallel. Their use has boosted scientific advancements in various fields such as biology, physics, medicine, and astronomy. However, scientific workflows are generally complex and have varied structures and characteristics that can have a significant impact on the result of a scheduling algorithm. Schedule a workflow consist to assign workflow tasks onto the resources of a computing infrastructure. Nowadays, the trend in information technology is the usage of Cloud computing environments to perform scientific workflow applications. However, cloud environments are experiencing a real problem of energy consumption. Inefficient resources management in cloud data centers has been identified as one of the main causes. That led to resources underutilisation, huge electricity bills, and reduction of the return of investment (ROI) for the cloud providers, and also high carbon dioxide emissions. As for the users, the respect of their defined deadline and budget is very important. In this thesis, we have proposed consecutively five workflow scheduling algorithms based on the structural properties of workflows. We have first investigated how to propose scheduling strategies to minimize both execution cost and execution time, which led to the proposal of two algorithms. Finally, we have investigated how to render our strategies more energy efficient. That led to the proposition of three scheduling algorithms aiming at minimizing the energy consumption, the execution cost, and the execution time. The three algorithms take advantage of the structural properties of the workflow as well as newly introduced scheduling concepts. At each step of our work, comparative simulations have been conducted between each of our proposals against state-of-the-art algorithms. Supported by adequate statistical tests, the analysis of the results reveals the levels of outperformance of our proposals both in the case of the two bi-objective algorithms than in the case of the three multi-objective ones aiming in addition at reducing energy consumption. The out-performance of the later ones in terms of energy-saving is established in 80% of types and workloads of workflows. Overall, one among the three, namely the Structure-based Multi-objective Workflow Scheduling with an Optimal instance type (SMWSO), is at least 50% more energy-saving, followed by our two other algorithms. As for the success rate, even though SMWSO scored overall the highest success rate, statistical tests revealed that there is no significant difference between our three algorithms and the baseline algorithm in terms of user satisfaction.
Content may be subject to copyright.