부산대학교 도서관

Automated composition and optimization of workflows in service-enriched environments is a challenging research area with strong implications in globally distributed systems such as Grid Computing and Cloud Computing. A workflow is composed of web services selected in accordance with user requirements. A strong formal realization of the problem is inevitable to ensure efficiency based on various interdependent parameters. We devise a mathematical model in order to map abstract workflows into concrete workflows satisfying user requirements represented by QoS parameters. Our model, which is based on the Multidimensional Multi-choice Knapsack Problem (MMKP), defines a happiness measure, that takes into account these requirements as well as the weights given to each requirement by the user. Then we develop a parallelizable branch and bound algorithm to maximize this happiness measure. We incorporate the Kepler Workflow tool, CORBA and C++ based optimization components to simulate two versions of the algorithm: a sequential version and a parallel version. We also indicate how to use heuristics to reuse the results of the parallel optimization under dynamic changes in requirements or service availability. Finally, we show a speedup analysis of our implementation.