Escalonamento de tarefas BoT em grades computacionais considerando o regime de incerteza

Abstract

Among the research challenges in Grid Computing we need to make the system robust both to the uncertainties of the different measures extracted from the computational infrastructure, and to the imprecision of the computations related to decision making. In particular, the appropriation of information relevant to the scheduling of tasks from a set of heterogeneous sources and dynamic behavior has been the focus of a significant number of studies. In this scenario, this masters dissertation aims to consolidate the fuzzy module fGrid for the scheduling of tasks in Grid Computing, aiming at its extension considering the information processing related to Computational Power and the Communication Cost of resources involved. For this, this approach employs type 2 fuzzy logic in the design for the so-called Int-fGrid extension, thus contributing to two relevant fronts, both with the uncertainty regime of this information and with the imprecision of the computations involved. In this sense, the work is based on a multi-valued approach, in particular integrating fuzzy logic and interval mathematics. As regards the evaluation of the modules, a simulation framework is used for distributed applications in order to facilitate the reproducibility of the different algorithms in different scenarios for grid infrastructure. The results obtained through the simulations with the schedules generated with the help of the fGrid and Int-fGrid modules show promise for the continuity of the study. These executions took into account different configurations of scenarios for grid infrastructures, while different application profiles made up of Bag-of-Tasks tasks showed that the approaches using fuzzy logic type 1 and type 2 obtained gains of up to 96.37% in makespan, compared to the Randon method of resource selection, 36.9 times compared to Round-Robin, and 18.5 times in relation to XSufferage.