Historically, data integration has centered around a data warehouse, with data ingest processes and algorithms that normalize data models, and queries that access this integrated data model.
Mashups, which are largely characterized by a lack of the data warehouse, must address the same concerns, matching of models from disparate data sources and careful access of the integrated set, however, the patterns are markedly different. Often, mashups are ad-hoc and may be used only once, however, the programming style is favorable even for solutions that will be in use for an extended period. In this alternate paradigm, where analysis is performed directly against the original data sources, interesting patterns emerge.
In this session we give an overview of the differences in style between traditional data integration and mashup approaches and we will study patterns such as resource composition, iteration, caching, filtering and error handling.