The NoSQL movement emerged from programmer aversion to strict physical data models and the widespread perception that such models do not scale to accommodate Big Data volumes. The result has been a headlong rush toward storage systems such as key-value stores. Such systems are sometimes called schema-less, which is misleading because when applied to structured-data phenomena, these systems do not eliminate schemas, but merely shift the burden for schema enforcement from the storage layer to the application layer. That shift brings along a cartload of troubles reminiscent of programming practice of the 1960′s, when every program had its own file format. To manage data in and extract information from a key-value store, you have to write code to walk the data structures.
The pendulum is swinging back. Key-value stores still have their place for truly schema-less phenomena, but for highly structured data such as geospatial data, genomic data, financial data and commercial data, there are significant benefits to honoring that structure in the storage layer — especially at Big Data volumes.
SciDB delivers those benefits. Through the array data model and its attendant query languages, SciDB provides ACID semantics and allows SQL-like (i.e., declarative) data manipulation and analysis. This is the cornerstone of ad-hoc, flexible, complex analytics — something that is not even theoretically possible with key-value stores.