The cardinality of the Cartesian product is n1 ?· n2 ?·??¦?· nn for the n restricted
dimensions. With these Cartesian product tuples, we perform a direct index access
on the composite index built on the fact table. For nonsparse fact tables and queries
that restrict most dimensions of the composite index in the order of the index
attributes, the access to the fact tuples is quite fast. However, for large sparse fact
tables and high dimensionality, such a query processing plan does not work effi-
ciently enough. The cardinality of the Cartesian product resulting from the dimension
predicates grows very fast, whereas the number of affected tuples in the fact
table may be relatively small. This is the point where a call is made for specialized
indexing or clustering methods.
Bitmapped join indices (O??™Neil & Graefe, 1995; O??™Neil & Quass, 1997) are often
used to speed up the access to the fact table. This type of star-join evaluation has
also been incorporated into a popular commercial system (Oracle, 2005). The so-
SELECT L.area, D.month, SUM(F.sales)
FROM SALES_FACT F, LOCATION L, DATE D, PRODUCT P
WHERE F.day = D.day AND F.store_id = L.store_id AND
F.product_id = P.item_id AND D.year = AND
L.population> 000000 AND P.category = ???air condition???
GROUP BY L.area, D.month
Figure 4. Example query
44 Karayannidis, Tsois, & Sellis
Copyright ?© 2007, Idea Group Inc.
Pages:
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289