Large-scale real-time distributed applications for real-time control are emerging in many domains including defense, space, financial markets, telecommunication, and industrial automation. To guarantee the performance of these applications operating in large IP internetworks, a scalable quality of service (QoS) architecture is needed. While considerable work has been done in real-time scheduling and bandwidth reservation schemes, the issues of scalability and feasibility of these schemes in actual networks remain largely unexplored. We propose a scalable QoS architecture in support of real-time systems, and focus on the implementation and testing of our proposed architecture using a network testbed. The key modules of our implementation are a real-time middleware, a graphical performance-monitoring tool, and a policy-based management software. We have integrated these applications with an existing Linux Traffic Control (TC) tool to deploy a scalable DiffServ-like QoS architecture. Our evaluation shows that the proposed architecture considerably improves performance of real-time applications as compared to the "best-effort" case.