Time-series database

Sensor data arrives as a stream of time-and-value points, often at high rates from many devices. Time-series databases compress it efficiently, make range queries ("the last 24 hours for unit 42") fast, and support downsampling and retention rules, such as keeping raw data for weeks and hourly averages for years. In industrial settings the equivalent is called a process historian.

Options include dedicated time-series databases, time-series extensions to general-purpose SQL databases, and managed cloud services. Data-modeling decisions (naming, tags, units, sample rates) matter more than the product choice and are hard to change later.