Optimizing the performance of a real-world wireless network is extremely challenging because of the difficulty to predict the network performance as a function of network parameters, and the prohibitively large problem size. The conventional network optimization approach relies on drive tests, trial-and-error and engineering experience, which is labor-intensive, error-prone and far from optimal. In this article, we introduce a new wireless network simulator, named simulated reality of communication networks (SRCON), that relies on a variety of white-box and machine-learning models to accurately reproduce the stochastic behavior (e.g., traffic arrival, user association, and signal propagation) of real-world 4G/5G mobile networks. We describe the key challenges, including the gigantic problem size, network coupling, stochasticity and many others, to develop SRCON. We also illustrate how SRCON can be used to optimize mobile network performance effectively. Notable performance gains were achieved after applying SRCON-enabled optimization to “conventionally-optimized” commercial wireless networks.