An Investigation into Dynamic Web Service Composition Using a Simulation Framework

Abstract

[Motivation] Web Services technology has emerged as a promising solution for creat- ing distributed systems with the potential to overcome the limitation of former distrib- uted system technologies. Web services provide a platform-independent framework that enables companies to run their business services over the internet. Therefore, many techniques and tools are being developed to create business to business/business to customer applications. In particular, researchers are exploring ways to build new services from existing services by dynamically composing services from a range of resources. [Aim] This thesis aims to identify the technologies and strategies cur- rently being explored for organising the dynamic composition of Web services, and to determine how extensively each of these has been demonstrated and assessed. In addition, the thesis will study the matchmaking and selection processes which are essential processes for Web service composition. [Research Method] We under- took a mapping study of empirical papers that had been published over the period 2000 to 2009. The aim of the mapping study was to identify the technologies and strategies currently being explored for organising the composition of Web services, and to determine how extensively each of these has been demonstrated and assessed. We then built a simulation framework to carry out some experiments on composition strategies. The rst experiment compared the results of a close replication of an ex- isting study with the original results in order to evaluate our close replication study. The simulation framework was then used to investigate the use of a QoS model for supporting the selection process, comparing this with the ranking technique in terms of their performance. [Results] The mapping study found 1172 papers that matched our search terms, from which 94 were classied as providing practical demonstration of ideas related to dynamic composition. We have analysed 68 of these in more detail. Only 29 provided a `formal' empirical evaluation. From these, we selected a `baseline' study to test our simulation model. Running the experiments using simulated data- sets have shown that in the rst experiment the results of the close replication study and the original study were similar in terms of their prole. In the second experiment, the results demonstrated that the QoS model was better than the ranking mechanism in terms of selecting a composite plan that has highest quality score. [Conclusions] No one approach to service composition seemed to meet all needs, but a number has been investigated more. The similarity between the results of the close replication and the original study showed the validity of our simulation framework and a proof that the results of the original study can be replicated. Using the simulation it was demonstrated that the performance of the QoS model was better than the ranking mechanism in terms of the overall quality for a selected plan. The overall objectives of this research are to develop a generic life-cycle model for Web service composition from a mapping study of the literature. This was then used to run simulations to replicate studies on matchmaking and compare selection methods.