Uma Abordagem para o Gerenciamento da Execução de Aplicações com Restrições de Tempo de Execução em Grades Computacionais Oportunistas

Abstract

An opportunistic grid computing environment takes advantage of idle computing cycles of regular computers and workstations that can be spread across several administrative domains for running high performance applications. Opportunistic grids are usually constructed from personal computers that do not need to be dedicated for executing grid applications. The grid workload must coexist with local applications executions, submitted by the nodes regular users. Thus, its execution environment is typically dynamic, heterogeneous and unpredictable failures occur frequently. In addition, the resources of an opportunistic grid can be used at any time for the execution of local tasks, making it difficult to preview the conclusion of the tasks running on the grid nodes. These characteristics hinder the successful execution of applications for which there are time restrictions related to its completion. This thesis presents a management mechanism specifically designed for opportunistic grid computing environments for handling the execution of applications with time deadlines set by users during their submission to the system. The proposed mechanism is based on a dynamic scheduling and rescheduling approach and was evaluated using a simulated model considering various typical scenarios of opportunistic grids. The results demonstrated the benefits of the proposed approach in comparison to traditional scheduling approaches applied in opportunistic grids.