Partitioned EDF scheduling for multiprocessors using a C=D task splitting scheme

An EDF-based task-splitting scheme for scheduling multiprocessor systems is presented in this paper. For m processors at most m −1 tasks are split. The first part of a split task is constrained to have a deadline equal to its computation time. The second part of the task then has the maximum time available to complete its execution on a different processor. The advantage of this scheme is that no special run-time mechanisms are required and the overheads are kept to a minimum. Analysis is developed that allows the parameters of the split tasks to be derived. This analysis is integrated into the QPA algorithm for testing the schedulability of any task set executing on a single processor under EDF. Evaluation of the C = D scheme is provided via a comparison with a fully partitioned scheme. Different heuristics for choosing the task to split are derived and evaluated. Issues pertaining to the implementation of the C = D scheme on Linux or via the Ada programming language are also discussed.