Development of computer system components in critical applications: problems, their origins and solutions

Abstract

The article is devoted to analysis of problems of the computer system development in the domain of critical applications. The main trends of this development were highlighted, which consisted in increased demands for performance based on parallelization of calculations, processing of approximate data and ensuring functional safety in accordance with the need for structuring for parallelism and fuzziness of the natural world, as well as with increased responsibility in decisions made. Analysis of problems encountered in implementation of existing solutions was carried out. There was a lag behind theories limited by the model of exact data from the practice of processing approximate data for modern systems receiving initial data from sensors, including safety-related systems. The problems of matrix structures, which underlie the design of modern computer systems and demonstrate low efficiency in performance and power consumption, as well as in providing functional safety, important for critical applications, are disclosed. The application of fault-tolerant solutions as the basis of functional safety and distrust of these solutions, which is manifested in the practice of using dangerous imitation modes, were noted. They recreate emergency conditions to improve the checkability in solving the problem of hidden faults, since a fault-tolerant solution does not become fail-safe when there is a shortage of checkability. An analysis was given to the sources of the problems considered and the possibilities of solving them from the point of view of a resource-based approach, which identifies the problem of hidden faults as a challenge of growth with a lag of components from the development of the system. The role of matrix structures in the backlog of components and the need to solve the problem by repeating the version redundancy for these structures are shown. Method of introduction of version redundancy into matrix structure on the basis of strongly connected versions for solution of problems of fault tolerance and checkability in complex is proposed. The effectiveness of the method is estimated on an example of the iterative array multiplier using its software model.