The full text of this thesis is available from the Newcastle University Library website: https://theses.ncl.ac.uk/dspace/handle/10443/1990
Waterworth, A., Department of Computing Science, University of Newcastle upon Tyne
The increased availability of sophisticated computer hardware and the corresponding decrease in its cost has led to a widespread growth in the use of computer systems for real-time plant and process control applications. Such applications typically place very high demands upon computer control systems and the development of appropriate control software for these application areas can present a number of problems not normally encountered in other applications. First of all, real-time applications must be correct in the time domain as well as the value domain: returning results which are not only correct but also delivered on time. Further, since the potential for catastrophic failures can be high in a process or plant control environment, many real-time applications also have to meet high reliability requirements. These requirements will typically be met by means of a combination of fault avoidance and fault tolerance techniques.
This thesis is intended to address some of the problems encountered in the provision of fault tolerance in real-time applications programs. Specifically, it consideres the use of replication to ensure the availability of services in real-time systems. In a real-time environment, provideing support for replicated services can introduce a number of problems. In particular, the scope for non-deterministic behaviour in real-time applications can be quite large and this can lead to difficulties in maintaining consistenet internal states across the members of a replica group. To tackle this problem, a model is proposed for fault tolerant real-time objects which not only allows such objects to perform application specific recovery operations and real-time processing activities such as event handling, but which also allows objects to be replicated. The architectural support reuired for such replicated objects is also discussed and, to conclude, the run-time overheads associated with the use of such replicated services are considered.