To cope with the requirements that ATM technology imposes on network management, it is necessary to discuss the changing role of network management. Currently, network management is concerned mainly with the support of isolating faults and detecting performance bottlenecks by long-term monitoring. However, having in mind the possibility to control the traffic in the network, the goal should be to prevent faults (i.e., unacceptable high cell losses), and to support efficient resource allocation in the ATM network.
The basic principle of ATM is statistical multiplexing, the success of which depends to a great extent on the statistical traffic behaviour in the network. It is therefore necessary to adjust resource allocation schemes with respect to certain characteristics of the traffic. Traditionally, network management is used to monitor the network, and the operator adjusts the network configuration after problems were detected.
A desired goal, however, is self-regulating network management, which should be able to configure the appropriate parts of the network with respect to the special requirements of the users before they recognize problems. These configuration decisions should be based on monitored traffic and knowledge of history network behaviour.
The necessity of a new role of ATM management has been recognized by various other authors. [8] introduces an architecture for the integration of performance management and resource control. Thereby providing the network operator with a set of control parameters to actively influence the way the Connection Admission Control (CAC) operates. The architecture they present is based on previous work of [5] and [6], and provides the possibility to adjust the CAC according to performance requirements. However, the choice of reasonable values for the control parameters is based on the experiences of the operator and the system acts on previously detected call request intensities. Unlike the approach presented in this paper, they do not consider any knowledge about usual (i.e., history-based) traffic requirements.
In another work of the same group, [1] presents an algorithm for virtual path configuration with the goal of minimizing call blocking and call setup times. So, this work aims in the prevention of problems, as well.
[4] also notes the necessity for management platforms to ''quickly determine and execute corrective actions'' in response to new technologies and services. They propose an intelligent multi-agent system where agents can work together and exchange knowledge about the current network state. As an example application of this system an 'early-cell-discard' approach is presented. There, the aim is to discard cells as early as possible, when congestion appearing further on the network would cause them to get lost anyway. On the contrary, the approach proposed in the present paper tries to minimize the occurrence of congestion itself.
[11] deals with the complex aspects of history management in general. This work presents an architecture to support the operator in the prediction of future behaviour of the network by providing him with a suitable environment ''to analyze the conditions leading to undesirable states'' in the past. However, the importance of combining history information and resource allocation objectives, especially in ATM networks, has not been considered up to now.
While recognizing the importance and necessity of the above approaches the work presented here takes a different route. The motivation is quite clear: As the success of a generous bandwidth sharing strategy in ATM depends on the statistical behaviour of the traffic - which certainly varies depending on providers, customers, available services and time - the base for management operations should be build on knowledge about this traffic.
In order to develop an adaptive algorithm, it is necessary to derive definite operations, that regulate the traffic in our network in order to match all user requirements best.
This work tries to identify such operations and the correct time to initiate them. A model is presented, which allows the derivation of management parameters to monitor and the actions to be carried out when certain values are exceeded in order to achieve the overall goal of 'Preventing Rather Repairing'.
To sketch the tasks to solve, it is necessary (i) to specify parameters to monitor the behaviour of the network, (ii) to specify thresholds for triggering the necessary actions, and (iii) of course, to specify the actions themselves. We present a model where these issues are tackled in order to achieve the goal of a self-regulating network management.
The paper proceeds as follows: Section 2 discusses network management tasks and introduces the 'connection-oriented view' for the definition of management goals. Afterwards, resource allocation is analyzed in section 3. Section 4 describes the proposed model and its main components: Specified parameters, actions, policies, and control mechanisms. In section 5, we discuss the implementational experiences gained. Section 6 concludes the paper and presents an outlook on future research.