MAG: uma grade computacional baseada em agentes móveis

Abstract

In recent years, Grid computing has emerged as a promising alternative to the integration and sharing of multi-institutional resources. However, constructing a Grid middleware is a complex task. Developers must address several design and implementation challenges, such as: e±cient management and allocation of distributed resources, dynamic task scheduling, high scalability and heterogeneity, fault tolerance, e±cient mechanisms for collaborative communication among Grid nodes, and security issues. MAG (Mobile Agents for Grid Computing Environments) explores the mobile agent technology as a way to overcome several of these challenges. MAG middleware executes Grid applications by dynamically loading the application code into a mobile agent. The MAG agent can be dynamically reallocated among Grid nodes through a transparent migration mechanism called MAG/Brakes, as a way to provide load balancing and support for non-dedicated nodes. MAG middleware also includes mechanisms for providing application fault tolerance, an essential characteristic for Grid environments. We make extensive use of the agent paradigm to design and implement MAG components, forming a multi-agent infrastructure for computational Grids. This master thesis describes MAG and MAG/Brakes architecture, implementation and performance issues.