Model-driven design of performance requirements

Obtaining the expected performance of a workflow is much simpler if the requirements for each of its tasks are well defined. However, most of the time, not all tasks have well-defined requirements, and these must be derived by hand. This can be an error-prone and time consuming process for complex workflows. In this work, we present an algorithm which can derive a time limit for each task in a workflow, using the available task and workflow expectations. The algorithm assigns the minimum time required by each task and distributes the slack according to the weights set by the user, while checking that the task and workflow expectations are consistent with each other. The algorithm avoids having to evaluate every path in the workflow by building its results incrementally over each edge. We have implemented the algorithm in a model handling language and tested it against a naive exhaustive algorithm which evaluates all paths. Our incremental algorithm reports equivalent results in much less time than the exhaustive algorithm.