Scientific Mashups: Runtime-Configurable Data Product Ensembles

Mashups are gaining popularity as a rapid-development, re-use-oriented programming model to replace monolithic, bottom-up application development. This programming style is attractive for the “long tail” of scientific data management applications, characterized by exploding data volumes, increasing requirements for data sharing and collaboration, but limited software engineering budgets. We observe that scientists already routinely construct a primitive, static form of mashup—an ensemble of related visualizations that convey a specific scientific message encoded as, e.g., a Powerpoint slide. Inspired by their ubiquity, we adopt these conventional data-product ensembles as a core model, endow them with interactivity, publish them online, and allow them to be repurposed at runtime by non-programmers. We observe that these scientific mashups must accommodate a wider audience than commerce-oriented and entertainment-oriented mashups. Collaborators, students (K12 through graduate), the public, and policy makers are all potential consumers, but each group has a different level of domain sophistication. We explore techniques for adapting one mashup for different audiences by attaching additional context, assigning defaults, and re-skinning component products. Existing mashup frameworks (and scientific workflow systems) emphasize an expressive “boxes-and-arrows” abstraction suitable for engineering individual products but overlook requirements for organizing products into synchronized ensembles or repurposing them for different audiences. In this paper, we articulate these requirements for scientific mashups, describe an architecture for composing mashups as interactive, reconfigurable, web-based, visualization-oriented data product ensembles, and report on an initial implementation in use at an Ocean Observatory.