On the Uses of Synchronization in Hard-real-time Systems

This paper presents an improved approach to the design of software for hard-real-time systems. The software for such systems is usually difficult to change because of constraints imposed by the need to meet absolute deadlines on processors with limited memory capacity. Dijkstra and others have shown that the concept of cooperating sequential processes can be used to improve the structure of software when concurrency is present. However, their techniques do not (1) deal with real-time deadlines, (2) are not efficient in their use of processor time and memory, and (3) do not allow simple efficient synchronization in the complex situations that arise in typical embedded systems. This paper presents a number of refinements to the basic concept that allow its use in hard-real-time embedded software. The three key ideas are, (1) a multi-level approach that allows convenient synchronization without a penalty in real-time response, (2) the use of a pre-run-time scheduler to allow more efficient use of resources, and (3) a new synchronization scheme that is both simple and general.