Bandwidth contracting in B-ISDN
It is anticipated that ATM/B-ISDN based telecommunication systems will support a wide variety of services, such as telephone, computer data communication, video communication, and control/alarm applications. The integration of greatly differing services in a single communication system poses new challenges for network bandwidth management and traffic control. This thesis addresses the problem of efficiently and effectively managing network bandwidth and performing traffic control in ATM/B-ISDN based networks. This study proposes a new dynamic traffic control algorithm, called the Bandwidth Contracting Algorithm (BCA), which allocates bandwidth to network users in a multi-level and flexible contracting strategy. Among the multiple types of contracts in BCA, Short-Term Green (STG) contract is of special interest, because it adds dynamic control by introducing bandwidth contracting at the burst-level. STG is intended for burst applications that require highly constrained cell loss but no bounds on delay and delay variation, such as the services in the nrt-VBR and ABR categories defined by the ATM Forum. This thesis concentrates on the performance of STG in two different network environments, LAN and WAN, through analytical modeling and simulation techniques. Also, performance comparisons are investigated between STG and rate-based flow control schemes suggested by the ATM Forum. This study shows that the burst-level control scheme STG is an efficient and effective traffic control scheme for loss-sensitive but delay-insensitive applications with burst traffic. Thus it is a useful alternative to the traffic control schemes for nrt-VBR and ABR services. Furthermore, this thesis develops a recursive approximation algorithm, fixed-point algorithm, to solve a class of non-product form queuing networks, which contains multiple customer chains, chain-dependent service times, multiple servers, and simultaneous resource possessions. This thesis shows that the fixed-point algorithm is advantageous with respect to the subnetwork algorithm of Souza e Silva and Muntz for the same class of queuing networks.