Data assimilation applications with high-dimensional numerical models show extreme requirements on computational resources. Thus, good scalability of the assimilation system is necessary to make these applications feasible. Sequential data assimilation methods based on ensemble forecasts, like ensemble-based Kalman filters and particle filters, provide such good scalability, because the forecast of each ensemble member can be performed independently. This parallelism has to be combined with the parallelization of both the numerical model and the data assimilation algorithm. While the filter algorithms can be implemented so that they are nearly independent from the model into which they assimilate observations, they need to be coupled to the numerical model. Using separate programs for the model and the data assimilation step coupled by disk files to exchange the model state information between model and ensemble data assimilation methods can be inefficient for high-dimensional models. More efficient is an online coupling strategy in which subroutine calls for the data assimilation are directly inserted into the model source code and augment the numerical model to become a data assimilative model. This strategy avoids model restarts as well as excessive writing of ensemble information into disk files and can hence lead to excellent computational scalability on supercomputers. The required modifications to the model code are very limited, such this strategy allows one to quickly extent a model to a data assimilation system. The online coupling is provided by the Parallel Data Assimilation Framework (PDAF, http://pdaf.awi.de), which is designed to simplify the implementation of scalable data assimilation systems based on existing numerical models. Further, it includes several optimized parallel filter algorithms. We will discuss the coupling strategy, features, and scalability of data assimilation systems based on PDAF.