On Connection Admission Control in Computer Networks

Doktorsavhandling, 2002

The best-effort nature of the Internet is a result of a relative simplicity of the network. This nature also allows the Internet to achieve high utilization of its resources. However, all these benefits are traded for inability of the Internet to provide a guaranteed or predictive quality of service (QoS). Therefore, the global computer network cannot reliably support real-time multimedia applications. It is a great challenge to turn the Internet into a QoS-capable communication network. Clearly, a QoS-capable Internet must remain simple and efficient. A QoS-capable network must contain three components: packet classification, flow isolation, and connection admission control (CAC). CAC performs admission decisions on connections generated by individual real-time network applications. A new connection is admitted to the network only if the admission does not deteriorate the QoS of connections that are already in progress and while delivering the target QoS to the new connection. Besides, CAC must ensure that maintaining QoS of connections in progress does not come at the cost of a low utilization of network resources. Efficient implementations of packet classification and flow isolation are widely available in contemporary network equipment. However, a simple admission control that would efficiently utilize network resources while having good scalability properties is still a challenge for the Internet research community. In this thesis the author presents a Heuristic-based Admission Control algorithm (HAC). Among properties of HAC is independence from the network infrastructure. This means that the algorithm does not require any additional protocols (e.g. ATM or RSVP). While being a per-hop CAC algorithm, complexity of the admission decision in HAC is very low. The algorithm has explicit configuration of the target QoS parameters, i.e. packet loss rate and queuing delay. In addition, HAC is capable of efficient self-tuning of its internal control parameters. Thanks to the fact that the algorithm is ignorant to traffic parameters of individual connections, it does not discriminate connections with certain values of traffic parameters as many CAC algorithms do. This was confirmed by simulation-based performance evaluation of the algorithm. The evaluation also revealed a good match between the target level of QoS and the actual level of QoS, together with high utilization of network capacity.