Arquitetura e reconfiguração dinâmica de recursos em clusters de servidores web utilizando sistemas multiagentes

Abstract

The resource allocation planning in a web server cluster is usually accomplished nowadays by the administrators of the computing environment based on different peak loads and the observation of the need of resources in previous situations. Once the Internet is quite dynamics as far as the use of resources is concerned, such task may be considered critical and inefiective if accomplished manually. This thesis describes the DARC architecture (Dynamic Reconfiguration Architecture for Clusters of Web Servers), an architecture for dynamic reconfiguration of servers clusters Web using a multiagent system. The agents carry out actions over the mechanism of QoS (Quality of Services) management in order to improve the eficiency of the use of available resources and they will be able to make dynamic decisions concerning the rate of utilization of resources and the load of the system. The multiagent system of DARC architecture will perform the necessary support to changes in available resources in the Web servers platform, without interference from the administrator. The presence of this would be needed in resource planning platform for Web servers, as it is he/she who has the initial knowledge for the allocation of resources. This thesis addressed the conception, specification, modeling Petri nets (Colored and Stochastic) and the implementation of this architecture in Java using RMI (Remote Method Invocation) for communication of agents. Several experiments and simulations which prove the efiectiveness of the proposed architecture are presented.