[cm277]

Well, an OLAP cube is a kind of a table/view. There are two core properties: the cube contains only one value ('column' in RDBMS-speak) and the data is defined across a set of 'dimensions' (levels in OLAP, unique indices with a foreign key in RDBMS).

The core insight though that the article is missing is that these dimensions are well-defined (and finite!) within the OLAP architecture and so are the roll-ups (aggregations) with their parent dimensions. The classical example is a date: a date has to roll-up to a week. Obvious. But the set of dates within the environment has to (or should anyway) also be well defined: so you can't just use any date, it has to fall, e.g. between 2019 and 2021.

Why? because the structure of the dimensions makes algebra across OLAP cubes trivial. In fact, OLAP algebra is pretty much linear algebra, just defined slightly differently for a more fluid problem space. That's why I like to think that OLAP isn't obsolete: in fact, most databases are just now becoming OLAP-y :)

Source: 20+ years of OLAP systems development.