The combination of shared bandwidth and rejection rate parameters, together with the quality of service predicted by neural networks in a novel strategy for connection admission control and call routing.The Broadband Integrated Service Digital Network (B-ISDN), which uses the Asynchronous Transfer Mode (ATM) technique based on fixed length packets called cells, requires complex flow control mechanisms to support narrow and broadband services.Since the communications networks based on synchronous time division multiplexing provide a quality of service independent of the load, the flow control mechanisms only need to reject the new connections when no resources are available. On the other hand, in packet based data networks, since the quality of service is load dependent, the flow control acts on the packet stream to guarantee a minimum quality of service. The B-ISDN requires a connection establishment phase within which the user and the network negotiate the quality of service of the connections which support the call, with some parameters being established on a probability basis, due to the competitive nature of the network. This implies that the control mechanisms must cover a wide variety of complex functions and the traditional analytical models become intractable.The application of neural networks and other artificial intelligence techniques is being recommended by many authors to implement a number of control functions in the B-ISDN. After the first work in this area presented in [1], many other publications in the last years have proposed feedforward or recurrent neural networks for applications such as connection admission control and call routing, service coding and traffic prediction, routing cells in spatial and temporal switches, policing functions of the usage parameters and selective cell discard, fault detection and network management. This paper discusses the use of neural networks for flow control applications. The simulation models of ATM traffic sources and B-ISDN components are briefly described and, based on these models, the traffic parameters are shown to be adequately predicted by neural networks. This capability is the basis of a novel technique for connection admission control and call routing which is also described. With this technique the admission decision is made according to the prediction of a few quality of service parameters expected for the new connections.
Why Neural Networks in B-ISDN Flow Control?The flow control functions in ATM networks include preventive and reactive actions. The connection admission control and the monitoring of traffic generated by users are examples of preventive actions, while the discarding of lower priority cells and some other congestion control mechanisms are included in the reactive actions.The B-ISDN control entities should also incorporate mechanisms for usage parameter control by policing the traffic generated by the calls at user interfaces and taking appro-1
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.