Monitoring systems for parallel distributed data management systems

• Autor(a) principal: Emanuel Augusto Santos Pinho
• Data de Publicação: 2016
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: https://repositorio-aberto.up.pt/handle/10216/89899
• topic: Engenharia electrotécnica, electrónica e informática; Electrical engineering, Electronic engineering, Information engineering

A BigData information system is usually composed by a large number of machines, which working closely in a network, perform distributed and parallel operations. Usually these systems go together with a monitor system that allows the control of many indicators, for instance, performance, resource allocation, error prevention, root-cause analysis. Nowadays, there are many systems which perform these functions, however, they have some disadvantages such as the graphics limitation, complex organization and its targeted approach to physical components.This study aims to understand what are the graphic visual mechanisms and interaction processes that allow the representation of performance monitoring data in grid computing environments, allowing the end-user information the ability to easily perceive the behavior of the BigData information system.This study is being carried in the broader context of a European project FP7 LeanBigData (which has already been graded with two excellents, in annual revision classifications) that aims to build a complete BigData information system, focusing its attention on solving efficiency problems. The monitor system innovative solution must be able to deal with the project needs, which are presented as a distinctive in problem-solving context.This study begins with an analysis of some of the systems used today. Based on their main qualities and disadvantages, it was formulated a hypothesis, conceived a solution and develop a proof of concept, where it was applied different techniques of representation of the resources structure, allowing the user interaction with a flow of use simple and intuitive. It was also devised an alternative approach to the representation of the physical components, designing a logical representation, which can be easily drawn conclusions on task organization of the network machines. To keep this interaction simple and complete, it requires a construction and edition of graphs and dashboards to enhance differentiation and manipulation capacities.With the implementation of the proof of concept proposed and approved by the project consortium, simulation tests were made with a simulation system carried out by the project.