Fault-Tolerant Dynamic Scheduling of Object-Based Real-Time Tasks in Multiprocessor Systems

Abstract

Multiprocessor systems are fast emerging as a powerful computing tool for real-time applications. The reliability required of real-time systems leads to the need for fault-tolerance in such systems. One way of achieving fault-tolerance is by Primary-Backup (PB) approach in which two copies of a task are run on two different processors. In this chapter, we compare and contrast three basic PB approaches – (i) primary-backup exclusive, (ii) primary-backup concurrent, and (iii) primary-backup overlapping – in the context of dynamic scheduling of object-based real-time tasks. The objective of this chapter is threefold: (a) to extend the PB-based fault-tolerant approaches, hitherto applied only to conventional real-time tasks, to object-based real-time tasks, (b) to compare these three approaches, in terms of schedulability and implementation complexity, and (c) to propose a dynamic scheduling algorithm for object-baaed real-time tasks, which can be used in conjunction with any of these PB-based fault-tolerant approaches. We have conducted extensive simulation studies to evaluate the performance of these three approaches, for tasks with resource and precedence constraints, for a variety of task and system parameters. Our simulation studies reveal some interesting results about the relative performance of these approaches.