Home > About Dimensional Database ... > About Cubes > Aggregation Strategies
![]() Previous |
![]() Next |
When an application issues a query, the OLAP engine either retrieves the stored summary values or calculates them on demand from a small number of stored data values. The result is a consistent response time across all areas of the cube. The target percentage identifies the number of aggregations to physically materialize at build time, leaving the remainder to be materialized transparently at query time. The goal is to find a setting that provides the optimal balance between build time and query response time. You can change the percentage at any time. The new percentage affects subsequent builds.
Note: Performance tests show that 40% is the optimal precompute percentage for most cubes. |
If you are using level-based aggregation, the best method for identifying levels for stored data is to determine the ratio of dimension members at each level, and to keep the ratio of members to be calculated at runtime at less than 10:1. This method assures that all answer sets can be returned quickly. Either the data is stored in the cube, or it can be calculated by rolling up 10 or fewer values into a single number. The time needed to roll up 10 values is trivial, and a well designed application limits return sets to an amount of data that an analyst can scrutinize easily. You can modify this ratio using your judgment on how frequently a level is accessed.