With the rapid development of cellular communication techniques, many recent studies have focused on improving the quality of service (QoS) in cellular networks. One characteristic of the systems in cellular networks, which can have direct impact on the system QoS, is the fluctuation of the system capacity. In this thesis, the QoS of systems with capacity fluctuations is studied from two perspectives: (1) priority queueing systems with preemption, and (2) the M/M/~C/~C system. In the first part, we propose two models with controlled preemption and analyze their performance in the context of a single reference cell that supports two kinds of traffic (new calls and handoff calls). The formulae for calculating the performance measures of interest (i.e., handoff call blocking probability, new call blocking and dropping probabilities) are developed, and the procedures for solving optimization problems for the optimal number of channels required for each proposed model are established. The proposed controlled preemption models are then compared to existing non-preemption and full preemption models from the following three perspectives: (i) channel utilization, (ii) low priority call (i.e., new calls) performance, and (iii) flexibility to meet various constraints. The results showed that the proposed controlled preemption models are the best models overall. In the second part, the loss system with stochastic capacity, denoted by M/M/~C/~C, is analyzed using the Markov regenerative process (MRGP) method. Three different distributions of capacity interchange times (exponential, gamma, and Pareto) and three different capacity variation patterns (skip-free, distance-based, and uniform-based) are considered. Analytic expressions are derived to calculate call blocking and dropping probabilities and are verified by call level simulations. Finally, numerical examples are provided to determine the impact of different distributions of capacity interchange times and different capacity variation patterns on system performance.

Ramp metering is an effective management strategy, which helps to keep traffic density below the critical value, preventing breakdowns and thus maintaining the full capacity of the freeway. Warrants for ramp metering installation have been developed by a number of states around the nation. These warrants are generally simple and are based on the traffic, geometry, and safety conditions in the immediate vicinity of each ramp (local conditions). However, advanced applications of ramp metering utilize system-based metering algorithms that involve metering a number of on-ramps to address system bottleneck locations. These algorithms have been proven to perform better compared to local ramp metering algorithms. This has created a disconnection between existing agency metering warrants to install the meters and the subsequent management and operations of the ramp metering. Moreover, the existing local warrants only consider recurrent conditions to justify ramp metering installation with no consideration of the benefits of metering during non-recurrent events such as incidents and adverse weather. This dissertation proposed a methodology to identify the ramps to meter based on system-wide recurrent and non-recurrent traffic conditions. The methodology incorporates the stochastic nature of the demand and capacity and the impacts of incidents and weather using Monte Carlo simulation and a ramp selection procedure based on a linear programming formulation. The results of the Monte Carlo simulation are demand and capacity values that are used as inputs to the linear programming formulation to identify the ramps to be metered for each of the Monte Carlo experiments. This method allows the identification of the minimum number of ramps that need to be metered to keep the flows below capacities on the freeway mainline segment, while keeping the on-ramp queues from spilling back to the upstream arterial street segments. The methodology can be used in conjunction with the existing local warrants to identify the ramps that need to be metered. In addition, it can be used in benefit-cost analyses of ramp metering deployments and associated decisions, such as which ramps to meter and when to activate in real-time. The methodology is extended to address incidents and rainfall events, which result in non-recurrent congestion. For this purpose, the impacts of non-recurrent events on capacity and demand distributions are incorporated in the methodology.