Federated learning (FL) is a machine learning system that allows a network of devices to train a model without centralized data. This characteristic makes FL an ideal choice for machine or deep learning using user data while maintaining privacy. Many applications, such as autonomous vehicles that adapt to pedestrian behavior and smart healthcare devices, require machine learning to be performed in edge computing environments. However, the performance of FL in edge computing environments has received limited research attention, and previous studies have not comprehensively evaluated heterogeneity aspects, such as system heterogeneity, statistical heterogeneity, and communication bandwidth. This study aims to fill this research gap by conducting experimental evaluations and detailed analyses of FL in edge computing environments. Specifically, we set up an experimental testbed based on the FL framework on the KubeEdge platform and evaluate the diverse effects of heterogeneity aspects on the model convergence time such as homogeneous versus heterogeneous hardware resources, balanced versus imbalanced data distributions, duplicate versus nonduplicate datasets, client participation ratio at each round, the client selection algorithm, and the communication bandwidth. Overall, this study contributes to the advancement of the field of FL by expanding the understanding of its performance in edge computing environments and providing guidelines for efficient and effective FL systems in heterogeneous environments, which can ultimately benefit various industries and domains.