For the initial column unique cardinalities (cucards), minimum values (cmin), and maximum values (cmax) of all of the TPC-D attributes, see Appendix C.4, Catalog Statistics for TPC-D.


Computation of Logical Properties

           The logical properties of new groups are computed by the function find_log_prop, defined for every logical operator. The description of this for every logical (bulk) operator is in Appendix L. However, a few general observations can be made.

           In the update of every equivalence class of the schema logical property by a logical operator the cucard3 but not cmin and cmax is updated. This is because the schema represents dynamic information and cucards change as operations are applied to a table. Although the cmin and cmax values would, in general, change (e.g. by a SELECT operator with a range predicate), this is not modeled in Model D. The initial cucard's and cmin and cmax (as calculated by the GET logical operator), would be determined by accessing the catalog using the ATTR_CAT entries of the schema.


Logical Item Properties

           Model D modeled two logical item properties: selectivity and free variables.

           The selectivity of a predicate is an estimate of the fraction of tuples for which the predicate will evaluate to TRUE. Recall that in order to determine the cardinality of a table produced by the SELECT logical operator, the selectivity of the predicate (second input taken by the SELECT operator) must be determined. The problem (addressed below) is how to compute the selectivity of these predicates.

3. In Model D the dynamically updated cucard is not used in selectivity estimates (see discussion below) but is used in calculating join cardinalities.

6.3 Properties:    <Logical Bulk> <Computation> <Item Props> <Selectivity> <Phys Prop>
  6.3: Log Prop  

 Page 7