This dissertation investigates various aspect of the design and testing of on-ramp metering control systems, including optimization-based control and microscopic freeway modeling. A new technique for generating optimal metering plans is developed. As with most predictive designs, the ramp metering rates are found as the solution to a nonlinear optimization problem. In contrast to previous designs, the new approach 1) produces a globally optimal solution to the nonlinear problem, 2) requires only to solve a single linear program, and 3) allows the enforcement of hard constraints on the on-ramp queue lengths. The price that is paid for these features is that the objective function being minimized is not Total Travel Time, but rather a member of a class of "TTT-like" objective functions. A TTT-like objective function is defined as a linear combination of mainline flows with weights that, similarly to the Total Travel Time cost weights, decrease linearly in time from some initial value to zero at the final time. An example application of the technique shows that the globally optimal metering plan with respect to a TTT-like objective function also performs well in terms of Total Travel Time. A macroscopic analysis of local traffic-responsive ramp metering on a short stretch of freeway, with a single on-ramp and no offramps, is also presented. The study compares the performance of two popular local traffic-responsive ramp metering algorithms: Alinea and Percent-Occupancy, and addresses issues pertaining to parameter tuning and loop-detector placement. The second half of the dissertation describes the construction of a detailed microsimulation model of a stretch of Interstate 210 in Pasadena, CA. The VISSIM microsimulation package was used to create this model. Descriptions of the data gathering and processing procedures, bottleneck identification, network coding, and model calibration are provided. The model is used to test the performance of candidate local traffic-responsive controllers. Questions concerning the relative merits of these controllers, parameter tuning, and loop-detector placement are addressed in the context of the large-scale microscopic model.