As networks evolve to carry multi-service traffic, each flow's bandwidth requirement is getting very diverse. We notice that even though the bandwidth usage level at a certain link is not high, if the number of connections carried on the link is excessive, it should be regarded as congested. We introduce connection resource notion into the traffic engineering. By incorporating the bandwidth resource notion and connection resource notion together into the problem formulation, we set up a more precise model for the multi-service network. To represent congestion cost at the links effectively, a nonlinear convex curve is used. The curve is then piece-wise linearized in the LP problem formulation. A calculation algorithm to get the link metric from the solution of the optimization problem is devised. In our framework, the congestion cost is caused either by the shortage of bandwidth or connection resource. Under the shortest path first routing scheme, we show through numerical experiments that the link metric set calculated by the new algorithm effectively reduces the congestion cost.

In voice trunking over ATM (VTOA) technology, the tandem is replaced by three components: trunk inter-working function (T-IWF), control and signaling inter-working function (CS-IWF), and ATM network. This new architecture calls for re-examination of the signaling channel delay issues, which are well-studied and well-specified for TDM networks. In VTOA, the ATM network together with the inter-working functions act as a virtual tandem switching system, which is called "ATM-based distributed virtual tandem switching system". The aim is that the performance of the ATM-based trunking network should be at least as good as the TDM-based network. The cross-office delay budget must be shared among the inter-working functions and the ATM network. Hence, the time for the ATM network to establish a switched virtual connection (SVC) is stringent. In this paper, we propose a simple adaptive SVC caching scheme to offload the high processing capacity burden on ATM switches. After a SVC is established in the usual way for a call, it is not torn down after the conversation is over. Instead, it is kept alive for variable duration (i.e., delayed release) with the expectation that there would be another call request for the same terminating end office during that time. The caching duration is adaptively changed with call arrival rate and call setup delay in the ATM network in order to stay within the delay budget specified by the requirements. The adaptability and the convergence of the algorithm are also shown by extensive simulations related to practical scenarios.

We present simple and informative new measurement functions. Their applications on real traffic traces provide insightful guidelines for measurement-based network control. We investigate practical scenarios and conclude that robust and insensitive measurement intervals can be found with the consideration of multiplexing and connection/flow dynamics. We find that a considerable percentage of the effective bandwidth is required by low frequency traffic in the context of realistic implementations. For a queuing node with the maximum allowable queuing delay d/sub max/, it is quantitatively shown that peak rate of the input traffic filtered (smoothed) with the averaging period of 200d/sub max/ captures more than 80% of the effective bandwidth when traffic aggregation and flow/connection dynamics are considered.