On the Scalability of Compositions of Service-Oriented Applications

One of the current challenges in the context of service-oriented applications is the generation of composition plans for applications that optimize their QoS attributes by taking advantage of the resources offered by different providers, and generate them as efficiently as possible. Scalability is indeed a major issue, and the problem becomes a real challenge for applications with big numbers of services. In this work, we propose a divide-and-conquer algorithm that exploits the architecture of applications, to reduce the size of the search space. With it, we are able to recompose the solution for the global problem, with a significant gain in execution time. A variant of the algorithm-in which, when a sub-problem cannot be further divided without loosing information, a solver is used to find the optimal solution for it-allows us to trade execution time and precision. We report on the extensive experimentation carried out, where applications with up to 2 000 services are considered, and which includes a comparison with the results delivered by GA solvers.