Index usage
Index usage
- Non-unique indexes must be used in combination to reduce result to a reasonable (if not unique) subset
- Prediction of selectivity for index 1, 2, ..., m used
seems easy, but values are very correlated, so that
either s << n (too many results) or s > n (none).
- So use some indexes for selection (assuring s < n ) and others for ranking (close match of index values)
s = 1/c1 * 1/c2 * … * 1/cm ~ n