Although service oriented architecture has reached a particular maturity level especially in the technological dimension, there is a lack of common and acceptable approach to design a software system through composition and integration of web services. In this thesis, a service oriented system design approach for Service Oriented Architecture based software development is introduced to fill this gap. This new methodology basically offers a procedural top-down decomposition of a given software system allowing several abstraction levels. At the higher levels of the decomposition, the system is divided into abstract nodes that correspond to process models in the decomposition tree. Any node is a process and keeps the sequence and the state information for the possible sub-processes in this decomposition tree. Nodes which are defined as process models may include some sub-nodes to present details for the intermediate levels of the model. Eventually at the leaf level, process models are decomposed into existing web services as the atomic units of system execution. All processes constructing the system decomposition tree are modeled with BPEL (Business Process Execution Language) to expose the algorithmic details of the design. This modeling technique is also supported with a graphical modeling language referred to as SOSEML (Service Oriented Software Engineering Modeling Language) that is also newly introduced in this thesis.

Current complex service systems are usually comprised of many other components which are often external services performing particular tasks. The quality of service (QoS) attributes such as availability, cost, response time are essential to determine usability and eciency of such system. Obviously, the QoS of such compound system is dependent on the QoS of its components. However, the QoS of each component is naturally unstable and di erent each time it is called due to many factors like network bandwidth, workload, hardware resource, etc. This will consequently make the QoS of the whole system be unstable. This uncertainty can be described and represented with probability distributions. This thesis presents an approach to calculate the QoS of the system when the probability distributions of QoS of each component are provided by service provider or derived from historical data, along with the structure of their compositions. In addition, an analyzer tool is implemented in order to predict the QoS of the given compositions and probability distributions following the proposed approach. The output of the analyzer can be used to predict the behavior of the system to be implemented and to make decisions based on the expected performance. The experimental evaluation shows that the estimation is reliable with a minimal and acceptable error measurement.

In this thesis, a service oriented decomposition approach: Use case Driven Service Oriented Architecture (UDSOA), is introduced to close the gap between business requirements and SOA (Service Oriented Architecture) design by including business use cases and system use cases into decomposition process. The approach is constructed upon Service Oriented Software Engineering (SOSE) modeling technique and aims to fill the deficits of it at the decomposition phase. Further, it aims to involve both business vision and Information Technologies concerns in the decomposition process. This approach starts with functional top-down decomposition of the domain. Then, business use cases are used for further decomposition because of their high-level view. This connects the business requirements and our SOA design. Also it raises the level of abstraction which allows us to focus on business services. Second step of the SOA approach uses system use cases to continue decomposition. System use cases help discovering technical web services and allocating them on the decomposition tree. Service oriented analysis also helps separating business and technical services in tightly coupled architecture conditions. Those two steps together bring quality in to both problem and solution domains.