The computing systems are becoming deeply embedded into ordinary life and interact with physical processes and events. They monitor the physical world with sensors and provide appropriate reaction and control over it. This cyber-physical interaction, which occurs through ubiquitous embedded systems, has the potential to transform how humans interact with and control the physical world. Applications of such systems include medical and healthcare systems, agile manufacturing, military applications, and environmental monitoring and control. For these applications, the demand for real-time data services is increasing since they are inherently data-intensive. Without real-time data service support, a significant amount of data coming from the physical world may not be properly handled in time, increasing the difficulty of the application development. However, providing real-time data services in such large-scale and geographically distributed environment is a challenging task. In particular, both unpredictable communicational delays and computational workloads of large-scale distributed systems can lead to large number of deadline misses. In this paper, we propose a real-time data service architecture called DRACON (Decentralized data Replication And CONtrol), which is designed to support large-scale distributed real-time applications. DRACON couples cluster-based replicasharing and a decentralized control structure to address communication and computational unpredictability, simultaneously. The cluster-based replica-sharing mechanism not only enables scalable and bounded-delay access to remote data with high probability, but also decouples clusters to have less interaction, allowing a decentralized, thus scalable, QoS control structure.