Network calculus is a theory developed to compute worst-case performance guarantees in networks. When analyzing complex networks (in particular multi-class), the analysis requires two-step approach: decomposing the network into several single-class networks, and analyzing these single-class network. In this talk, I will show that some scheduling policies (such as DRR) exhibit some inter-dependency structures and require that these two steps are intricate. I will present an iterative scheme that improves both the performance bounds and stability region compared to the state of the art.