Managing Service Dependencies in Service Compositions

In the Internet of Services (IoS) providers and consumers of services engage in business interactions on service marketplaces. Provisioning and consumption of services are regulated by service level agreements (SLA), which are negotiated between providers and consumers. Trading composite services requires the providers to manage the SLAs that are negotiated with the providers of atomic services and the consumers of the composition. The management of SLAs involves the negotiation and renegotiation of SLAs as well as their monitoring during service provisioning. The complexity of this task arises due to the fact that dependencies exist between the different services in a composition. Dependencies between services occur because the complex task of a composition is distributed between atomic services. Thus, the successful provisioning of the composite service depends on its atomic building blocks. At the same time, atomic services depend on other atomic services, e.g. because of data or resource requirements, or time relationships. These dependencies need to be considered for the management of composite service SLAs. This thesis aims at developing a management approach for dependencies between services in service compositions to support SLA management. Information about service dependencies is not explicitly available. Instead it is implicitly contained in the workflow description of a composite service, the negotiated SLAs of the composite service, and as application domain knowledge of experts, which makes the handling of this information more complex. Thus, the dependency management approach needs to capture this dependency information in an explicit way. The dependency information is then used to support SLA management in three ways. First of all dependency information is used during SLA negotiation the to ensure that the different SLAs enable the successful collaboration of the services to achieve the composite service goal. Secondly, during SLA renegotiation dependency information is used to determine which effects the renegotiation has on other SLAs. Finally, dependency information is used during SLA monitoring to determine the effects of detected violations on other services. Based on a literature study and two use cases from the logistics and healthcare domains different types of dependencies were analyzed and classified. The results from this analysis were used as a basis for the development of an approach to analyze and represent dependency information according to the different dependency properties. Furthermore, a lifecycle and architecture for managing dependency information was developed. In an iterative approach the different artifacts were implemented, tested based on two use cases, and refined according to the test results Finally, the prototype was evaluated with regard to detailed test cases and performance measurements were executed. The resulting dependency management approach has four main contributions. Firstly, it represents a holistic approach for managing service dependencies with regard to composite SLA management. It extends existing work by supporting the handling of dependencies between atomic services as well as atomic and composite services at design time and during service provisioning. Secondly, a semi-automatic approach to capturing dependency information is provided. It helps to achieve a higher degree of automation as compared to other approaches. Thirdly, a metamodel for representing dependency information for SLA management is shown. Dependency information is kept separately from SLA information to achieve a better separation of concerns. This facilitates the utilization of the dependency management functionality with different SLA management approaches. Fourthly, a dependency management architecture is presented. The design of the architecture ensures that the components can be integrated with different SLA management approaches. The test case based evaluation of the dependency management approach showed its feasibility and correct functioning in two different application domains. Furthermore, the performance evaluation showed that the automated dependency management tasks are executed within the range of milliseconds for both use cases. The dependency management approach is suited to support the different SLA management tasks. It supports the work of composite service providers by facilitating the SLA management of complex service compositions