Toward a Methodology for the Development of Network Management Solutions: a Study on Mobility, Autonomy and Distribution in Agents

Resumo

Current computer networks require an intricate management activity in order to provide high quality of service to system us ers. Usual automation tools follow a client-s erver app ro ach, which centralizes the processing. Such management kind lacks flexibility and fault tolerance and generates high network traffic. A promising alternative approach is based on intelligent agents capable of taking decisions more autonomously and migrating between devices, allowing a more distributed management. This work discusses the impact that properties such as mobility, autonomy and distribution in agents have in the development of management solutions for corporative networks. Considering some metrics (processing time, CPU consumption, space savings ), we have implemented and made an evaluation of different multiagent architectures varying the levels of these properties. As case study we used the disk space management problem in UNIX/NFS machines. A simulator was developed to carry out experiments with higher control. The results can be used as a guide toward a methodology for construction of management solutions based on mobile, autonomous and distributed agents.