
172 Automated Physical Database Design and Tuning
The finest and coarsest granularities give us two points of reference on
performance vs. number of distinct values. The exact shape of this curve
cannot be easily determined but in practice can be modeled as a smooth
logarithmic relationship. We can then perform a curve-fitting procedure
that gives as a formula to obtain, for a given number of distinct values,
the expected benefit of using such granularity for the corresponding
dimension column.
3. Define candidate granularities for each dimension column. Once we have
identified the finest granularity value, 4 to 10 progressively coarser gran-
ularities are ...