Operating system principles and constructs for dynamic multi-processor real-time control systems

This work addresses on-line multiprocessor scheduling for hard real-time control systems. The example application used is an autonomous mobile robot however, the results are applicable to other domains with dynamic, real-time control requirements such as agile manufacturing and virtual reality.; Contemporary techniques for implementing real-time control systems include parallelized application tasks, object oriented system design and hybrid architectures. These techniques require new interfaces too, and make greater demands of the task scheduler. This work defines Rapid, a software architecture for dynamic, concurrent scheduling of hard real-time tasks on multiprocessors where the scheduling requests consist of sets of tasks. This architecture (1) is flexible to accommodate complex, changing control requirements, (2) improves performance over previous non-concurrent scheduling and, (3) provides new application interfaces for scheduling sets of tasks. This work also defines a new technique for accommodating unpredictable tasks in hard real-time systems. This technique has broader application than previous techniques. The architecture is evaluated on an example scheduler developed for the Kendall Square Supercomputer scheduling a simulated workload and embedded in a schema-based mobile robot simulation.