Most research papers should have one thing in common: a clear and expressive evaluation of proposed solutions to problems. However, evaluating solutions is interestingly a challenging task: when using human-constructed examples or real-world data, it is difficult to assess to which degree the data represents the input spectrum also of future demands. Moreover, evaluations which fail to show generalization might hide algorithm weak-spots, which could eventually lead to reliability and security issues later on. To solve this problem we propose Toxin, a framework for automated, data-driven benchmarking of, e.g., network algorithms. In a first proof-of-concept implementation, we use Toxin to generate challenging traffic datasets for a data center networking use case.