In hierarchically clustered fact tables, this
translates to one or more simple multidimensional range queries on the underlying
multidimensional structure that is used to store the fact table data (operator
MD_Range_Access in Figure 5(a)). Moreover, since data are physically clustered
according to the hierarchies and the ranges originate from hierarchical
restrictions, this will result in a low-I/O evaluation of the range selection.
??? Step 2. Computing necessary joins: The tuples resulting from the fact table
contain the h-surrogate values and the measure values. At this stage, there
might be a need for joining these tuples with a number of dimension tables in
order to retrieve certain hierarchical or feature attributes that the user wants to
have in the final result and might also be needed for the grouping operation.
We call these joins residual joins. Note that all these join operations (the Residual_
Join nodes in Figure 5(a)) are equi-joins on key-foreign key attributes
and therefore each fact table tuple is joined with exactly one dimension table
tuple.
??? Step 3. Performing grouping, filtering, and ordering: Finally, the resulting
tuples may be grouped and aggregated and the groups further filtered and
ordered for delivering the result to the user. The Group_Select operator in
Figure 5(a) performs these actions.
46 Karayannidis, Tsois, & Sellis
Copyright ?© 2007, Idea Group Inc.
Pages:
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293