A Study On Composite Service Selection Method Based On Service Correlation Patterns

Service-oriented architecture is a style of software architecture which reuses and combines loosely coupled services to build, maintain and integrate applications in order to improve productivity throughout the software life cycle. Each concrete service encapsulates information resource and function of application component. Workflow specifies the logic relationships among services. A service level agreement (SLA) defines mutual understandings and expectations of a service between service providers and service consumers. Each of abstract services in the workflow should be bound concrete services to satisfy the QoS constraints imposed by the SLA.In practical scenario, there may be existing correlations among Web services. Therefore, when determing the binding between abstract service and concrete service, it also needs to consider other abstract services. Current researches assumed the correlations among services have already existed or service providers have declared the correlations in the services specification. At the application level, these researches focused on establishing the correlated QoS model, and selecting services based on the model. However, these methods face some problems in practical environment. Firstly, the reasons for correlations among services are complicated. It is difficult to directly analyze which services have correlations. This will negatively affect the application of correlations in composite service. Secondly, the similar functionalities of different applications can be accomplished by the same QoS correlation services. Current researches ignored how to use the correlation services as the reusable knowledge which applied to other applications. Moreover, these researches only considered how to make an initial selection in correlation context. They also ignored how to reselect services when exceptions occurred. Especially, there existing correlations between completed and uncompleted services. In the final, these methods only focus on the single SLA-aware service selection problem, and ignore that composite service instances should provide multiple QoS levels for different users in the cloud environment.In order to solve above problems, this dissertation proposes a composite service selection method based on service correlation patterns. Data mining technology is introduced to analyze the QoS correlation services according to the execution and invocation information, and to establish the knowledge about business model of composite service. The knowledge is the basis of business modeling for composite service, and the QoS correlation services can improve the quality of single and multiple SLA-aware selection method. Surrounding this idea, in-depth studies are done in the following aspects.(1) For the problem that the correlation services are difficult to analyze and how to make the correlation services as effective and reusable knowledge, this dissertation proposes a service correlation patterns extracting method. This method mines the services with better effectiveness when they are used together. The mined services will be regarded as QoS correlation services. Then the control flow will be inferred according to the service invocation events log. In the final, the correlated abstract services are extracted. The correlation services compose the candidate service set of this kind of abstract service. Correlated abstract services, its corresponding correlation services, the control flow associated with correlated abstract services will be recorded as service correlation patterns. Abstract services and control flow respectively reflect the domain knowledge about decomposition of business activities according to the requirements and establishing the logic relations among these activities. The patterns can be reused in different applications to solve the same problem without reinventing the wheel. The better practical effect can be acquired through using the correlation services that are frequently tested.(2) For the problem that select the correlation patterns to establish the composite service workflow according to the functional requirements, this dissertation proposes a pattern selection method based on matching input and output parameters. The parameters of input and output for correlated abstract services are matched with input and output of users. The full-matching or partial-matching abstract services will be selected. The choosen abstract services and its associated control flow can be regarded as domain knowledge to establish workflow. Their corresponding correlation services are used to support the SLA-aware service selection.(3) For the problem that how to select services based on service correlation patterns and reselect to avoid the SLA violation, this dissertation proposes a service selection method supporting correlation QoS and a reselection method taken into account the situation that the completed services are correlated with uncompleted ones. This method firstly presents the correlated QoS model and the service selection method supporting the correlated QoS model. Then, in the reselection method, the abstract services in the remaining workflow will be classified as different types. The candidate service sets of abstract services with different types will be determined. Based on the updated candidate service set, the reselection method will be completed.(4) For the problem that multiple SLA-aware composite service selection method supporting correlated QoS, this dissertation proposes a composite service selection method supporting service sharing across multi-SLAs in the correlation context. Service instances are regarded as sharing resources, and they can be accessed simultaneously by multiple compositions with different quality levels. This dissertation also gives QoS aggregation function based on correlated QoS. This method avoids the problem that the feasible solutions cannot be found for some users caused by several high performance services are monopolized by other users. And it also enhances the utilization ratio of service instances.