In safety-critical systems (such as nuclear power plants, automated robot control systems, automatic landing systems for aircraft, etc.) tasks not only have to meet deadlines, but most of these are critical in the sense that the system would not survive in case of a certain number of deadline failures of subsequent task instances. In such a critical stage, a task instance is said to have become essentially critical. However, beyond this special real-time responsiveness (here the successful handling of essentially critical task instances that we have termed survivability, safety-critical systems must also satisfy rigid dependability requirements (reliability and fault tolerance). In order to meet as many of these conflicting requirements, in environments that are typically unpredictable, a very high amount of adaptability of system functions is demanded. Safety-critical systems have gained rapidly increasing relevance in research and development, both industrial and commercial, and are typically distributed. All this makes even the design of large safety-critical systems an extraordinarily complex modeling and engineering challenge. In the Melody project a distributed real-time operating system has been developed that is tailored to the requirements of safety-critical systems. It was started at Wayne State University in Detroit, Michigan, and is now in the phase of an operational distributed lab prototype in our labs.