A parallel approach for user-centered QoS-aware services composition in the Internet of Things

Abstract

The Internet of Things (IoT) refers to an infrastructure of interconnected smart devices that aim to provide various services. The proliferation of IoT objects and devices offering functionally equivalent services but differing in their quality of service (QoS) levels makes the issue of services composition one of the biggest challenges for the service computing community. Various evolutionary-based approaches have been proposed in the literature to find sub-optimal service compositions in a reasonable computation time. However, most of these approaches have high composition time and/or a limited composition quality as they rely on a sequential exploration of the composition search space using a fixed size population. To address these limitations, a parallel differential evolution-based approach with population size reduction for QoS-aware service composition (PDE-QSC) is proposed in this paper. Unlike existing evolutionary-based approaches, the proposed approach is characterized by a parallel exploration of the composition space through a population size reduction strategy. Specifically, in this approach, the composition population is divided into two sub-populations. To reduce the composition time and improve the quality of the composition, the composition sub-populations evolve simultaneously using different evolution processes and are then merged to form a single population, thus increasing the population diversity. To further improve the performance in terms of composition time and composition quality, a linear reduction strategy is proposed to adaptively reduce the size of the composition population by eliminating compositions that do not meet the QoS requirements. Simulations based on real datasets demonstrate the superiority of the PDE-QSC approach over five baseline approaches and its suitability for large-scale IoT environments.