Abstract
(English)
|
The increased demand for new services on the one hand and the need to reduce costs on the other hand, increasingly forces Telecom Operators and other service providers to
improve the management of resources. One possibility to achieve cost reductions by better utilization of resources lies in distributed control which is tightly coupled to agent technology. Research in the area of intelligent agents has shown that a society of independently acting agents, in which agents negotiate with each other about resources, may achieve promising resource utilization results. For our purposes, we found the negotiating agents who are given the task of managing a certain kind of resources are a very interesting paradigm. Intelligence is then often just an add-on for some special agents, while the actual power comes from the negotiating and management facilities of the agents.
The IMPACT project was implementing a prototype system for controlling ATM Networks where Control-plane tasks such as Admission Control, Routing, Connection Handling, Virtual Path capacity management, etc. will be performed by higher-intelligence, distributed software entities called 'Intelligent Agents'.. In the IMPACT system, the Intelligent Agents constituted the control infrastructure of the ATM network. The Agents will interact by means of a high-level communication language (the 'Agent Communication Language' -ACL) in order to decide whether new connection requests can be admitted, how new connections will be routed, what resources will be allocated to them, etc.
The advantages of performing network control by using tntelligent Agents can be basically summarised in two points:
* Flexible management of resources and, moreover, in the way desired by the network operator or the service provider, rather than the way that the equipment vendor has embedded in the network elements (this can be achieved since the agents' behaviour can be customised and may override the internal control behaviour of the
1
network elements).
* Control functionality that is not feasible otherwise, such as reactivity to unexpected events of various forms or re-allocation of resources allocated to existing connections in order to optimise utilisation.
IMPACT demonstrated the above advantages of agent-based network control, by implementing the IMPACT system on an ATM testbed with heterogeneous equipment. This implementation was not meant to result in a commercial, full-fledged network control system (coping with every aspect of network and service control); it was mostly a prototype demonstrating the mentioned benefits from the use of Intelligent Agents.
The work used the EXPERT platform in Basel. The application concentrated on Connection Admission Control (CAC) but to demonstrate the attractiveness of agents in controlling co-operating functions, the work also introduced an accounting/charging agent.
|
