Kinds of database

Relational databases are not the only way to store data. They are brilliant for structured, consistent data, but struggle when the data is huge, varies from record to record, or is mostly about relationships. Other database models were invented for those jobs, and you need to know what each is for.

NoSQL (“not only SQL”) databases do not use fixed, linked tables. They have flexible schemas and are designed to scale out across many servers, so they suit big, varied, fast-changing data such as web applications and real-time analytics. There are four main families.

• Document databases store each record as a self-contained document, usually JSON, holding fields and values (and even nested data) together. No joins are needed to read one, and each document can have different fields. Good for content management and product catalogues; the risk is repeated data and inconsistency. Example: MongoDB.
• Key-value databases store a value under a key, like a giant dictionary. Very fast to read and write, used for caching and session data, but you can only look things up by key. Example: Redis.
• Wide-column databases store data in tables with rows and flexible columns: different rows can have different columns. They handle huge, sparse data sets and spread easily across servers but are complex to design. Example: Cassandra.
• Graph databases store nodes (people, places, things) and edges (the relationships between them). They are excellent for complex connections such as social networks and recommendation engines, but need a specialised query language.

• Cloud databases run on a cloud platform. They can be relational or NoSQL. They scale up and down on demand, can be spread over data centres across the world for faster access, and are billed by use. You can run one yourself on a rented virtual machine (full control, more work) or use a fully managed service where the provider handles backups, patching and scaling.
• Spatial databases are built to store and query geographic data: points, lines and polygons (a shop, a road, a country border). They power maps and geographic information systems (GIS), and add questions such as “which of these points are within 5 km of here?”.
• In-memory databases keep all their data in RAM instead of on disk. Reads and writes are extremely fast, so they suit real-time work such as gaming, sensor data and trading. The limits are the cost and size of RAM, and the need to protect the data from a power failure.