Structured data

Structured data is data that is organized according to a predefined schema, so that every record conforms to a known shape. The schema fixes the fields, their data types, and the relationships between them before any data is written, and the store enforces it on every insert and update. The classic form is tabular data, arranged into rows and columns, but the defining trait is the up-front, machine-readable schema rather than any particular storage shape.

Structured data is one of three conventional classifications of digital data, alongside semi-structured data, which carries a self-describing shape that can vary from record to record, and unstructured data, which has no predefined schema at all. The three sit on a spectrum rather than behind hard borders, but the structured end is the one a machine can parse without first having to infer the shape.

Where it lives

Structured data is the native form of the relational database, which stores records as rows in tables linked by foreign keys and queried with SQL. A database more generally is an organized collection of structured or semi-structured data, so the classification lines up with the most mature and widely deployed storage family. Spreadsheets, CSV files, and fixed-width legacy formats carry the same shape outside a database engine.

The schema is what unlocks the guarantees relational systems are built around. Because every column has a known type and every row obeys it, the store can validate input, enforce data integrity and referential integrity, and offer the ACID transactional guarantees that ad-hoc data shapes cannot. It can also build indexes on columns it knows to exist, which is what makes structured data cheap to search by predicate.

Trade-offs

The price of an up-front schema is rigidity. Adding a field means altering the schema, migrating existing rows, and updating every system that reads them. Data that does not fit the columns — free text, nested records, images — either has to be shoehorned into a blob column or pushed out to a less structured store. For workloads that change shape often, or that mix many formats, the schema-on-read model of NoSQL systems is usually the better fit, and the big data ecosystem is built around platforms that accept structured, semi-structured, and unstructured data side by side.

The flip side of that flexibility is that the structured store’s guarantees are lost. Validation moves to the application, indexes and query planning get harder, and the optimizer can no longer rely on a known column shape. The choice between structured and less-structured storage is a trade between integrity and agility, not a question of one being strictly better.

See also

References