Motivation

For business data analysis, it is often necessary to aggregate data across many dimension (attributes). For example, a table Transaction with attributes Product (P), Customer (C), Date (D) and Sales (S). An analyst could query for finding :

• sum of sales by P, C.
• sum of sales by D, C.
• sum of sales by P.

To make interactive analysis, speed is critical. So many of these aggregates are precomputed.

Precomputation speed is also important as it influences how frequently the aggregates are updated.