State-of-the-art database management systems are inappropriate for real-time applications due to their lack of speed and predictability of response. To combat these problems, the scheduler needs to be able to take advantage of the vast quantity of semantic and timing information that is typically available in such systems. Furthermore, to improve predictability of response, the system should be capable of providing a partial, but correct, response in a timely manner. We therefore propose to develop a semantics for real-time database systems that incorporates temporal knowledge of data-objects, their validity, and computation using their values. This temporal knowledge should include not just historical information but future knowledge of when to expect values to appear. This semantics will be used to develop a notion of approximate or partial computation, and to develop schedulers appropriate for real-time transactions.