Dimensionality Reduction of Quality Objectives for Web Services Design Modularization

Abstract

With the increasing use of service-oriented Architecture (SOA) in new software development, there is a growing and urgent need to improve current practice in service-oriented design. To improve the design of Web services, the search for Web services interface modularization solutions deals, in general, with a large set of conflicting quality metrics. Deciding about which and how the quality metrics are used to evaluate generated solutions are always left to the designer. Some of these objectives could be correlated or conflicting. In this paper, we propose a dimensionality reduction approach based on Non-dominated Sorting Genetic Algorithm (NSGA-II) to address the Web services re-modularization problem. Our approach aims at finding the best-reduced set of objectives (e.g. quality metrics) that can generate near optimal Web services modularization solutions to fix quality issues in Web services interface. The algorithm starts with a large number of interface design quality metrics as objectives (e.g. coupling, cohesion, number of ports, number of port types, and number of antipatterns) that are reduced based on the nonlinear correlation information entropy (NCIE).The statistical analysis of our results, based on a set of 22 real world Web services provided by Amazon and Yahoo, confirms that our dimensionality reduction Web services interface modularization approach reduced significantly the number of objectives on several case studies to a minimum of 2 objectives and performed significantly better than the state-of-the-art modularization techniques in terms of generating well-designed Web services interface for users.