Hybrid: unifying centralised and distributed network management using intelligent agents

Abstract

Centralised network management platforms suffer from an information processing bottleneck which establishes a trade-off between the size of a network and the precision with which it can be managed. High speed networking protocols such as asynchronous transfer mode (ATM) networks exacerbate this problem as they generate large amounts of data and require more sophisticated traffic control policies than traditional circuit-switched networks. Recognition of the need to retain the benefits of centralised management has led researchers to identify categories of telecommunications management functions which are suited to distribution. This paper describes the hybrid system which applies distributed artificial intelligence (DAI) intelligent agent concepts and technologies to unify the benefits of centralised and distributed management. DAI promotes the use of behavioural rather than functional control. We argue that this is an important abstraction tool for constructing, maintaining and understanding large and complex systems. In particular, in order to improve network management scalability, we propose a hierarchical system of independent controllers (agents) with local problem-solving and decision-making capabilities. Each agent acts much like existing centralised management systems, but collectively they communicate to maintain system-wide objectives and to resolve conflicts of interest between themselves. The hybrid architecture has wide application potential for distributed management and control tasks, particularly interdomain network management, service management and manager of managers (MOM).