Fault-tolerance for real-time multiprocessor operating systems

Most fault-tolerant systems rely on special hardware features to implement their recovery schemes. These special hardware features usually require specialized software programming and development techniques. Entirely software based systems often require specialized programming constructs to control the system's fault-tolerant features. This tends to make the software machine specific and therefore not portable to new machine architectures or even to updated versions of the same execution environment.;Real-time systems tend to run on unique hardware platforms and these reflect themselves in the system's synchronization and I/O paradigms. Software components developed for one system are rarely transportable to another system.;There is a requirement for a real-time operating system supporting fault tolerance capable of being implemented on many standard machines of varying architectures. It is also desirable that the operating system make a maximum use of resources in the absence of system failures.;Hardware can be freed for other work if the state of software components can be restored, in the event of a processor failure, rather than duplicated through active redundancy.;The objective of this project is to modify a real-time operating system for multiprocessor applications to support automatic system recovery from the failure of any one processor unit in the system. The support does not require the use of specialized operating system calls or force changes in programming style. The implementation is being done without specialized fault-tolerant hardware. The National Research Council of Canada's Harmony real-time operating system (12) is the foundation for this work. (Abstract shortened by UMI.).