Optimizing Resource Allocation for Scientific Workflows Using Advance Reservations

Scientific applications are more and more faced with very large volumes of data and complex, resource-intensive workflows that process or analyze these data. The recent interest in web services and service-oriented architectures has strongly facilitated the development of individual workflow activities as well as their composition and the distributed execution of complete workflows. However, in many applications concurrent scientific workflows may be served by multiple competing providers, with each of them offering only limited resources. At the same time, these workflows need to be executed in a predictable manner, with dedicated Quality of Service guarantees. In this paper, we introduce an approach to Advance Resource Reservation for service-oriented complex scientific workflows that optimize resource consumption based on user-defined criteria (e.g., cost or time). It exploits optimization techniques using genetic algorithms for finding optimal or near-optimal allocations in a distributed system. The approach takes into account the locality of services and in particular enforces constraints imposed by control or data flow dependencies within workflows. Finally, we provide a comprehensive evaluation of the effectiveness of the proposed approach.