Optimal structure of fault-tolerant software systems

• Published in: Reliability engineering & system safety Vol. 89; no. 3; pp. 286 - 295
• Main Author: Levitin, Gregory
• Format: Journal Article
• Language: English
• Published: Oxford Elsevier Ltd 01.09.2005; Elsevier
• Subjects: Applied sciences; Computer science; control theory; systems; Exact sciences and technology; Expected execution time; Fault-tolerant programming; Genetic algorithm; Operational research and scientific management; Operational research. Management science; Optimization; Reliability; Reliability theory. Replacement problems; Software; Software engineering; Fault-tolerant programming; Performance; Reliability; Genetic algorithm; Expected execution time; Optimization; Probabilistic approach; Generating function; Program execution; Software component; System reliability; Recovery; Fault tolerance; Resource constraint; Version management; Fault tolerant system; Execution time
• ISSN: 0951-8320; 1879-0836
• DOI: 10.1016/j.ress.2004.09.001

This paper considers software systems consisting of fault-tolerant components. These components are built from functionally equivalent but independently developed versions characterized by different reliability and execution time. Because of hardware resource constraints, the number of versions that can run simultaneously is limited. The expected system execution time and its reliability (defined as probability of obtaining the correct output within a specified time) strictly depend on parameters of software versions and sequence of their execution. The system structure optimization problem is formulated in which one has to choose software versions for each component and find the sequence of their execution in order to achieve the greatest system reliability subject to cost constraints. The versions are to be chosen from a list of available products. Each version is characterized by its reliability, execution time and cost. The suggested optimization procedure is based on an algorithm for determining system execution time distribution that uses the moment generating function approach and on the genetic algorithm. Both N-version programming and the recovery block scheme are considered within a universal model. Illustrated example is presented.