Abstract
Future dynamic distributed hard real-time systems may control unpredictable environments and will need predictable and flexible runtime systems that can handle unknown and changing task populations. In this extreme case, task scheduling must be purely dynamic accepting any type of task, and hard deadlines must be dynamically guarded against transient over-load conditions. This paper addresses the design and implementation of a pure dynamic distributed task scheduler for hard real-time systems. The algorithms have been proven to guarantee task deadlines even during transient overloads. Its task model is quite unrestrictive, supporting dynamically scheduled periodic and aperiodic tasks both of which may have arbitrary release times and deadlines. Moreover, resource scheduling confiders precedence constraints and communication requirements. Finally, the schedulability test has been proven correct and has a complexity of O(n). A domain-wide scheduler is described which maintains its predicted maximum response time even when local nodes are under heavy load. Experimental results confirm the proven expectations of the scheduling scheme.