Understanding Aggregations

As we begin to understand the implications of cube processing options, let’s explore a bit more about the concept of an OLAP aggregation. What exactly is an aggregation? It’s a pre-aggregated, (usually summed) stored value. Remember that data is loaded into the cube from the rows in the fact table. These rows are loaded into the source fact table from the various source systems at the level of granularity (or detail) defined in the grain statements. For example, it’s typical to load the fact rows at some time granularity. For some clients, we’ve loaded at the day level—that is, sales per day; for others, we’ve loaded at the minute level.

In some ways, an OLAP aggregation is similar to an index of a calculated column of a relational table—that is, the index causes the results of the calculations to be stored on disk, rather than the engine having to calculate them each time they are requested. The difference, of course, is that OLAP cubes are multidimensional. So another way to think of an aggregation is as a stored, saved intersection of aggregated fact table values. For the purposes of processing, aggregations are considered data (rather than metadata). So, the data in a cube includes the source fact table rows and any defined aggregations.