Designing distributed real-time systems is an intricate and not very well-structured task. Besides complex functional capabilities, these systems often require high standards on real-time performance, reliability, and availability. Though architectures like CORBA hide the distributed infrastructure of the underlying platform from the application software, supporting real-time performance and fault-tolerant services remains a complex design problem.This paper presents an architecture for distributed real-time systems in which fault-tolerance is supported by transparent replication of mission-critical components. The architecture is based on a shared data space through which components interact. The architecture is supported by middleware that provides a distributed implementation of the shared data space. The implementation is based on a light weight asynchronous communication protocol that is able to deliver real-time performance.