Interactivity in scientific workflows: a survey

The analysis of scientific data is inherently exploratory, which leads to iterative refinements of the design of the workflows supporting such analysis. That is, scientists test certain operator configurations; they debug the implementation of an analysis pipeline based on intermediate results; and they formulate hypotheses on the characteristics of the data as well as the outcomes, which are then assessed through the execution of the workflow. While common workflow systems provide various means to support scientists in the design of scientific workflows, they traditionally emphasize reuse, reproducibility, and traceability. Using these systems, the iterative refinement of a workflow is inefficient and ineffective. It involves changing and restarting a workflow as a whole in order to explore design choices and there is no support to systematically assess the impact of design choices. Against this background, first attempts to provide support for interactive workflow execution have been made recently. However, these initial proposals vary widely in their aims and addressed use cases, in their operationalization, and in their integration into existing frameworks. In this chapter, we set out to structure the emerging field of interactive scientific workflows. To this end, we present a taxonomy for interaction support in scientific workflows and adopt it for a systematic survey of existing concepts as well as their integration into workflow systems. We thereby provide an overview of the state of the art in interactive scientific workflows and carve out directions for future research.