Enforcing Advance Reservations for E-Science Workflows in Service Oriented Architectures

Scientific Workflows have become an important tool to perform complex calculations, especially when individual operations are made available as services in Service Oriented Architectures. At the same time, Quality-of-Service aspects and Advance Reservation of resources by means of Service Level Agreements (SLA) are topics that get ever-increasing attention in order to make best use of available resources in a predictable manner. The support of such SLAs at the level of workflows raises two interrelated issues pertaining (i) to the temporal prediction of reservation start time and duration of individual activities, and (ii) to the actual enforcement of resource commitments at the provider side.In this paper, we outline our vision of a distributed workflow engine with support for SLAs and Advance Reservations. We focus on reservations addressing processing capabilities, i.e., shares of CPU power. In particular, we present a module of the system that is responsible for the enforcement of such reservations at the individual service providers' nodes, which, by means of a Fuzzy Controller adjusting task priorities, makes sure that the SLAs are met in a fair way.