Traditional sensor network deployments consisted of fixed infrastructures and were relatively small in size. More and more, we see the deployment of ad-hoc sensor networks with heterogeneous devices on a larger scale, posing new challenges for device management and query processing. In this paper, we present our design and prototype implementation of XSense, an architecture supporting metadata and query services for an underlying large scale dynamic P2P sensor network. We cluster sensor devices into manageable groupings to optimise the query process and automatically locate appropriate clusters based on keyword abstraction from queries. We present experimental analysis to show the benefits of our approach and demonstrate improved query performance and scalability.