Article Preview
Top1. Introduction
Cloud computing is a collection of accessible resources that can be dynamically composed based on the user’s requirements. Cloud services are defined and provided at three levels: Software as a Service (SaaS), Platform as a Service (PaaS), and Infrastructure as a Service (IaaS) (Gutierrez-Garcia & Sim, 2012). This arrangement allows users of cloud services to focus on what the services provide to them rather than how the services are implemented or hosted. The number of cloud providers is increasing (e.g., GoGrid, Amazon, and Google), along with the number of services they offer (e.g., Software-as-a-Service applications and computing resources). The increasing demands for complex cloud services create the need for dynamic, automated, and adaptive composition of cloud services in a decentralized open environment in which cloud services can dynamically arrive and depart (Gutierrez-Garcia & Sim, 2012).
According to Navimipour and Vakili (2017), existing approaches to web/cloud service composition can be divided into three distinct categories, namely, framework-based, agent-based, and heuristic-based approaches.The advantage of framework-based approaches to service composition is that they organize and manage searching, selecting, and composing the cloud services (Navimipour & Vakili, 2017). However, the framework-based approaches presented in this paper (Benfenatki, Da Silva, Benharkat, & Ghodous, 2014; Di Martino, Cretella, & Esposito, 2016; Nguyen, Lelli, Papazoglou, & van den Heuvel, 2012; Qui, 2014; Sasikaladevi, 2016; Tsai, Sun, & Balasooriya, 2010; Zhou, Athukorala, Gilman, Riekki, & Ylianttila, 2012) are defined in a centralized manner, which can lead to system overloads. The agent-based approaches that have been proposed (Bastia, Parhi, Pattanayak, & Patra, 2015; Gutierrez-Garcia & Sim, 2012; Mellah, Hassas, & Drias, 2013; Rodrigues, Leitão, & Oliveira, 2014; Sim, 2012; Singh, Juneja, & Malhotra, 2015; Val, Rebollo, Vasirani, & Fernandez, 2014) are very flexible and autonomous (Kumar, 2012; Talia, 2011). However, they suffer from the problem of communication through messages and do not adapt well to large and open environments, where services can dynamically arrive and depart and the QoS may improve or deteriorate. Mostafa et al. (2014) demonstrated the effectiveness of a stigmergy mechanism that was used to manage the dynamic nature of trust, but this work is based exclusively on the criteria of trust and reputation, which creates a difficulty for selecting new members joining the system for the first time, as they have no historical record. Additionally, the proposed adaptation is only linked to changes in the QoS and not to changes in the environmental structure (Mostafa, Zhang, & Bai, 2014). Finally, the heuristic-based approaches that have been proposed (Gohain & Paul, 2016; Karimi, Isazadeh, & Rahmani, 2016; Li, Jiang, & Ge, 2014; Seghir & Khababa, 2016; Wu, Chen, & Huang, 2016; Yu, Chen, & Li, 2015) incorporate optimization algorithms in order to produce optimized composition plans. However, these approaches suffer from a high level of complexity.
The purpose of cloud service composition is to fulfill functional and/or nonfunctional user requirements. Functional requirements concern the overall result of the application that is to be developed, while nonfunctional requirements (quality of service; QoS) concern the quality of the composition, such as the response time, availability, reliability, and cost. As the number of cloud services offering similar functionalities increases, identification of the cloud service composition plan with the best quality becomes a critical problem (Mostafa et al., 2014). This has led to proposals of effective methods for selecting the best service composition plan.