Recent studies reveal that the main cause of the well-known unfairness problem in wireless networks is the ineffective coordination of CSMA-based random access due to topology asymmetry. In this paper, we take a modeling-based approach to understand and solve the unfairness problem. Compared to existing works, we advance the state of the art in two important ways. First, we propose an analytical model called the G-Model, which accurately characterizes the ineffective coordination of medium access in asymmetrical topologies. The G-Model can estimate network performance under arbitrary parameter configurations. Second, while previous works decompose a wireless network into embedded basic asymmetric topologies and study each basic topology separately, we go beyond the basic asymmetrical topology and design a model-driven optimization method called Flow Level Adjusting (FLA) to solve the unfairness problem for larger wireless networks. Through extensive simulations, we validate the proposed G-Model and show that FLA can greatly improve the overall fairness of wireless networks in which basic asymmetric topologies are embedded.