Note that in order to push a residual join operation above a grouping operation the
join must be carefully modified. This can be done either by grouping the dimension
table and modifying the join condition or by using special join algorithms that join
each fact table tuple with only one tuple (the first matching tuple) from the dimension
table. This is because all initial residual joins are equi-joins on the key attribute.
The details of the hierarchical pregrouping transformation and its definition as
an algorithm appear in Karayannidis et al. (2002) and Tsois (2005). A theoretical
analysis of the transformation, its generalization as well as a proof of correctness
can be found in Tsois (2005) and Tsois and Sellis (2003).
Future.Trends
Speculating about the future trends in data warehouse query processing in general,
we believe that there are two main factors that will drive the processing requirements
in the near future:
1. Continuously increasing data volumes that one needs to analyze.
2. Continuously increasing rates by which data for analysis are generated on the
one hand and increasing need for up to date information on the other.
The first factor calls for extremely scalable storage organizations that exploit a
plethora of successful techniques such as semantic based physical data clustering,
precomputation of aggregates, and fast indexing.
Pages:
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306