Re: Storage for multiple variable-length attributes in a single row

Dear all

When ingesting mobility (IoT) data into MobilityDB

https://mobilitydb.com/

we transform very wide (2K attributes) car mobility data of high frequence (every tenth of a second) from flat format (e.g. CSV) into MobilityDB format in which there is a single record per trip and each of the signals is transformed into a temporal attribute (tbool, tint, tfloat, ttext, tgeompoint, tgeogpoint), which are temporal extensions of the corresponding PostgreSQL/PostGIS base types (bool, int, float, text, geometry, geography). All temporal types are stored using extended format, e.g.,

    CREATE TYPE tfloat (

      internallength = variable,
      [...]
      storage = extended,
      alignment = double,
      [...]
    );

Given that each temporal value can be very wide (on average 30K timestamped  points/floats/text/... per trip) our first question is

* Is extended the right storage for this ?

Our second question is how all the 2K temporal attributes are stored, which may be

* on a single table space

* in one table space per attribute

which in other words, relates to the question row vs column storage.

Many thanks for your insight

Esteban