Performance analysis of bandwidth allocation strategy with state-dependent bernoulli access

Abstract

The main objective of this dissertation work is to study the performances of a bandwidth allocation strategy with state-dependent Bernoulli access for traffic controls in wideband integrated networks, and to optimize the control parameter of Bernoulli access probability in the strategy. Integrated networks provide a common transmission facility which has a large amount of bandwidths to serve heterogeneous traffics of different characteristics. In order to allocate the available bandwidth to the various types of traffics in an efficient, flexible and fair manner, a bandwidth allocation strategy (i.e., static and dynamic allocation) is required. For this purpose, we first studied a static bandwidth allocation strategy with state-dependent Bernoulli access (SDBA) for integration of wideband (WB) and narrowband (NB) traffics. In order to protect overloading of one traffic and to guarantee grade-of-service requirements for each type of traffics, the strategy has a number of features such as preemptive priority, specification of a cut off parameter and a movable boundary scheme, and utilizes the system states which consist of the amount of bandwidths occupied by the WB traffic and the number of NB messages waiting in the buffer. We applied the strategy in two environments. In the first environment, a channel capacity is divided by basic bandwidth units. In the second environment, instead of quantizing a channel capacity, the total bandwidth is divided into three regions. Several explicit formulas of the blocking probability of WB traffic and the mean waiting time of NB traffic were derived by using the matrix geometric method under the assumption that WB traffic is blockable and NB traffic is queueable. By numerical examples, we showed that this strategy shows better performance than other strategies which do not use the SDBA property, and can easily be adapted to a varying traffic load by changing the parameters of the cut off parameter and the queue thresholds. Nex...