A Method for Consistent Non-Local Configuration of Component Interfaces
Service-oriented computing is a popular technology that facilitates the development of large-scale distributed systems. However, the modular composition and flexible coordination of such applications still remains challenging for the following reasons: 1) the services are provided as loosely coupled black boxes that only expose their interfaces to the environment; 2) interacting services are not usually known in advance: web services are dynamically chosen to fulfil certain roles and are often replaced by services with a similar functionality; 3) the nature of the service-based application is decentralised. Loose coupling of web services is often lost when it comes to the construction of an application from services. The reason is that the object-oriented paradigm, which is widely used in the implementation of web services, does not provide a mechanism for service interface self-tuning. As a result, it negatively impacts upon the interoperability of web services. In this dissertation we present a formal method for automatic service configuration in the presence of subtyping, polymorphism, and flow inheritance. This is a challenging problem. On the one hand, the interface description language must be flexible enough to maintain service compatibility in various contexts without any modification to the service itself. On the other hand, the composition of interfaces in a distributed environment must be provably consistent. Our method is based on constraint satisfaction and Boolean satisfiability. First, we define a language for specifying service interfaces in a generic form, which is compatible with a variety of contexts. The language provides support for parametric polymorphism, Boolean variables, which are used to control dependencies between any elements of interface collections, and flow inheritance using extensible records and variants. We implemented the method as a constraint satisfaction solver. In addition to this, we present a protocol for interface configuration. It specifies a sequence of steps that leads to the generation of context-specific service libraries from generic services. Furthermore, we developed a toolchain that performs a complete interface configuration for services written in C++. We integrated support for flexible interface objects (i.e. objects that can be transferred in the application along with their structural description). Although the protocol relies solely on interfaces and does not take behaviour concerns into account, it is capable of finding discrepancies between input and output interfaces for simple stateful services, which only perform message synchronisation. Two running examples (a three buyers use-case and an image processing application) are used along the way to illustrate our approach. Our results seem to be useful for service providers that run their services in the cloud. The reason is twofold. Firstly, interfaces and the code behind them can be generic as long as they are sufficiently configurable. No communication between service designers is necessary in order to ensure consistency in the design. Instead, the interface correspondence in the application is ensured by the constraint satisfaction algorithm, which we have already designed. Secondly, the configuration and compilation of every service are separated from the rest of the application. This prevents source code leaks in proprietary software which is running in the cloud.