Multiple Start Modifications of Ant Colony Algorithm for Multiversion Software Design

• Published in: Advances in Swarm Intelligence Vol. 11655; pp. 191 - 201
• Main Authors: Saramud, Mikhail V., Kovalev, Igor V., Losev, Vasiliy V., Voroshilova, Anna A.
• Format: Book Chapter
• Language: English
• Published: Switzerland Springer International Publishing AG 2019; Springer International Publishing
• Series: Lecture Notes in Computer Science
• Subjects: Ant algorithm; Architecture; Multiversion programming; Optimization; Reliability; Software design
• ISBN: 9783030263683; 3030263681
• ISSN: 0302-9743; 1611-3349
• DOI: 10.1007/978-3-030-26369-0_18

The paper discusses the use of an optimization algorithm based on the behaviour of the ant colony to solve the problem of forming the composition of a multiversion fault-tolerant software package. A model for constructing a graph for the implementation of the ant algorithm for the selected task is proposed. The modifications of the basic algorithm for both the ascending and the descending design styles of software systems are given. When optimizing for downstream design, cost, reliability, and evaluation of the successful implementation of each version with the specified characteristics are taken into account. When optimizing for up-stream design, reliability and resource intensity indicators are taken into account, as there is a selection from already implemented software modules. A method is proposed for increasing the efficiency of the ant algorithm, which consists in launching a group of “test” ants, choosing the best solution from this group and further calculating on the basis of it. A software system that implements both modifications of the basic ant algorithm for both design styles, as well as the possibility of applying the proposed multiple start technique to both modifications, is considered. The results of calculations obtained using the proposed software tool are considered. The results confirm the applicability of ant algorithms to the problem of forming a multiversion software package, and show the effectiveness of the proposed method.