In this paper, we address the data management aspect of large-scale pervasive computing systems. We aim at building an infrastructure that simultaneously supports many kinds of context-aware applications, ranging from room level up to nation level. This allembracing approach gives rise to synergetic benefits like data reuse and sensor sharing. We identify major classes of context data and detail on their characteristics relevant for efficiently managing large amounts of it. Based on that, we argue that for large-scale systems it is beneficial to have special-purpose servers that are optimized for managing a certain class of context data. In the Nexus project we have implemented five servers for different classes of context data and a very flexible federation middleware integrating all these servers. For each of them, we highlight in which way the requirements of the targeted class of data are tackled and discuss our experiences.