<p>Web service composition has become a promising technique to build powerful enterprise applications by making use of distributed services with different functions. In the age of big data, more and more web services are created to deal with a large amount of data, which are called data-intensive services. Due to the explosion in the volume of data, providing efficient approaches to composing data-intensive services will become more and more important in the field of service-oriented computing. Meanwhile, as numerous web services have been emerging to offer identical or similar functionality on the Internet, web service composition is usually performed with end-to-end Quality of Service (QoS) properties which are adopted to describe the non-functional properties (e.g., response time, execution cost, reliability, etc.) of a web service. In addition, the executions of composite web services are typically coordinated by a centralized workflow engine. As a result, the centralized execution paradigm suffers from inefficient communication and a single point of failure. This is particularly problematic in the context of data-intensive processes. To that end, more decentralized and flexible execution paradigms are required for the execution of data-intensive applications. From a computational point of view, the problems of QoS-aware data-intensive web service composition and execution can be characterised as complex, large-scale, constrained and multi-objective optimization problems. Therefore, genetic programming (GP) based solutions are presented in this thesis to address the problems. A series of simulation experiments are provided to demonstrate the performance of the proposed approaches, and the empirical observations are also described in this thesis. Firstly, we propose a hybrid approach that integrates the local search procedure of tabu search into the global search process of GP to solving the problem of QoS-aware data-intensive web service composition. A mathematical model is developed for considering the mass data transmission across different component services in a data-intensive service composition. The experimental results show that our proposed approach can provide better performance than the standard GP approach and two traditional optimization methods. Next, a many-objective evolutionary approach is proposed for tackling the QoS-aware data-intensive service composition problem having more than three competing quality objectives. In this approach, the original search space of the problem is reduced before a recently developed many-objective optimization algorithm, NSGA-III, is adopted to solve the many-objective optimization problem. The experimental results demonstrate the effectiveness of our approach, as well as its superiority than existing single-objective and multi-objective approaches. Finally, a GP-based approach to partitioning a composite data-intensive service for decentralized execution is put forth in this thesis. Similar to the first problem, a mathematical model is developed for estimating the communication overhead inside a partition and across the partitions. The data and control dependencies in the original composite web service can be properly preserved in the deployment topology generated by our approach. Compared with two existing heuristic algorithms, the proposed approach exhibits better scalability and it is more suitable for large-scale partitioning problems.</p>
Experimental Analysis for Semantic based Large Scale Service Composition using Deep Learning
