Engineering holistic fault tolerance

Abstract

Fault-tolerant software should be engineered to be maintainable as well as efficient with regards to performance and resources. These characteristics should be evaluated before deployment of the software. However, the main focus is very often made on the functional features of the application, whereas fault tolerance mechanisms are neglected. As a result, they are often neither maintainable nor efficient. The concept of Holistic Fault Tolerance was introduced to deal with these issues. It is a novel crosscutting approach to the design and implementation of fault tolerance mechanisms for developing reliable software applications that meet non-functional requirements, such as performance and resource utilisation. The thesis starts with the description of problems that were motivating for the idea of Holistic Fault Tolerance. These problems are related to resource utilisation requirements of modern computer-based systems, since more resources like hardware components and energy are required to process modern computational tasks and ensure performance and reliability of the computation. Moreover, the complexity of these systems grows, leading to maintainability deterioration, especially of those system parts, which are responsible for satisfying non-functional requirements, such as reliability, performance and resource usage. After analysis of the problems and motivations, the engineering approach to Holistic Fault Tolerance is introduced and main engineering steps are defined. Next, an architectural pattern for Holistic Fault Tolerance is presented. The method to refine the proposed architecture and ensure efficiency of a particular system under development is demonstrated during the modelling step. Then the implementation of Holistic Fault Tolerance based on the proposed architecture and modelling is described in detail. Finally, the Holistic Fault Tolerance architecture is evaluated with regards to efficiency and maintainability. The evaluation demonstrates that Holistic Fault Tolerance assists in meeting the non-functional requirements, makes fault tolerance mechanisms easier to maintain and ensures higher modularity of the source code