Uma Contribuição para Avaliação de Custos do Uso de Nuvens Computacionais no Suporte a Aplicações Científicas

Abstract

This Thesis discusses the adoption of cloud infrastructures to support the execu tion of scientific applications in academia, where it is common to share computational resources among different research groups. For the proper management of shared resources, the computational cloud environment appears promising in terms of universal access and better use of financial resources. The general objective of this work is to rationalize, in relation to financial costs, the adoption of cloud computing for academic research support. To do this, it is necessary to determine the costs involved in implementing a solution, serving a federated community of educational and research institutions. In this context, alternatives with public, private, and hybrid cloud environments are addressed, identifying the advantages and disadvantages of each, as well as their respective financial impacts. Different usage scenarios for cloud infrastructures were conceived in order to combine options of federated and public clouds with scaling strategies, including cloud bursting. As a scientific contribution, this work brings the finding that public clouds can be used to support a federated cloud and the development of a simulator for network environments, the WCSim (Workflow Cloud Simulator), in which applications modeled as workflows could be used from tasks that relate to a bag-of-tasks. To evaluate the behavior of workflow executions on a federated cloud, along with the cloud bursting technique, case studies were applied for use in WCSim. The results of the experiments allow us to verify that the spot pricing modality should be considered in the evaluations of possible cloud service provider contracts. It is also observed that intra and inter-site load balancing can be well explored, enhancing performance gains