A distributed data flow model for composing software services

This thesis presents a distributed data-flow model for composing autonomous software services, as might be provided over the web. The autonomous services are linked to form a data processing system, controlled by one node, which we call the megaservice. The distributed data-flow model allows direct data exchange among the autonomous services. This is different from the traditional centralized data-flow model where the megaservice is the central hub for all the data traffic. A theoretical analysis shows that the distributed data-flow model has better performance and scalability than the centralized data-flow model. The distribution of data communications fully utilizes the network capacity among the autonomous services, and avoids bottlenecks at the megaservice.; A prototype infrastructure for service composition, the Flow-based Infrastructure for Composing Autonomous Services (FICAS), has been implemented to support the distributed data-flow model. FICAS is a collection of software modules that support the construction of autonomous services, facilitate the specification of the megaservice, and enable the efficient execution of the megaservice. The distribution of data communications is enabled by a metamodel defined for autonomous services, which separates the data interchange from the control processing in the services. Autonomous services conforming to the metamodel can be coordinated by a centralized controller, while data communications are distributed among the services.; Data transformations and similar computational tasks are often needed to interface autonomous services. Since in the distributed data-flow model the data do not flow through the megaservice, such transformations have to be carried out externally. To achieve that we define mobile classes, dynamic processing routines that can be loaded onto an autonomous service to prepare data local to the service. By moving computations closer to data, the amount of data traffic can be significantly reduced for a megaservice, hence improving the performance of the megaservice.; Based on FICAS, an engineering service infrastructure is constructed for project management applications in the construction industry. The infrastructure demonstrates that the distributed data-flow model is suitable for composing large-scale software services.