Supporting distributed real-time applications in largescale and heterogeneous distributed environments has attracted much attention recently. However, the decentralized and dynamic characteristics of the systems present a number of challenges in managing the processor and networking resources and scheduling the application execution across multiple peers. In this paper, we address these problems with a novel resource management architecture for scheduling soft real-time applications in overlay networks. Key to our approach is that nodes maintain a partial view of resource availability of their peers in local profiles, keep track of the urgency of the requests and balance application execution across multiple nodes. Our techniques are entirely decentralized and use only local knowledge considering unpredictable latencies and changing resource availability.