Full text
| |
| Author(s): |
Gregorio Baggio Tramontina
Total Authors: 1
|
| Document type: | Doctoral Thesis |
| Press: | Campinas, SP. |
| Institution: | Universidade Estadual de Campinas (UNICAMP). Instituto de Computação |
| Defense date: | 2008-10-10 |
| Examining board members: |
Jacques Wainer;
Autran Macedo;
Jose Palazzo Moreira de Oliveira;
Claudia Maria Bauzer Medeiros;
Flávio Keidi Miyazawa
|
| Advisor: | Jacques Wainer |
| Abstract | |
Workflow systems are present in today's companies to automate and optimize their business processes. One activity of these systems is to direct the execution of its tasks to their executors, and when there is an excess of these tasks, a decision must be made regarding the order of execution of these tasks. Current workflow systems use the FIFO (First In First Out) policy to make those decisions, executing the tasks according to their arrival order. But it is possible to achieve measureable gains by doing this task ordering differently. This work presents a study of the behavior of scheduling techniques in workflow systems and proposes a methodology to apply these techniques to more complex workflow scenarios. To build this methodology, the authors analyzed the behavior of selected scheduling techniques in basic workflow scenarios. Both local and global scheduling techniques were chosen, ranging from dispatching rules to a genetic algorithm. The studied workflow scenarios have three important characteristics: they are dynamic, they have uncertainties on the processing time of the tasks in the process, and they also have uncertainties on the route each task follows within its process. To handle these uncertainties, this work uses the guess and solve technique, proposed by the authors in previous works. Simulation experiments were performed to generate the results on the behavior of the scheduling techniques in the basic workflow scenarios, and the mentioned results were analyzed numerically and statistically using ANOVA. The work then proposes a methodology to apply the best scheduling techniques in subcomponents of complex scenarios. The authors applied simulation to the complex scenarios using the proposed methodology, and the results show that it brings improvements when compared to the other tested techniques. When there is no parallel execution in the process, the worst case of the methodology is that it is no worse then the others, and its general case brings measureable improvements. When there is parallel execution, the methodology has its results deteriorated, which points to future research in the field. (AU) | |