The h-surrogates should be system assigned and maintained attributes, and typically
they should be made transparent to the user. The actual implementation of the h-surrogates
depends heavily on the underlying physical organization of the fact table.
Proposals for physical organizations (Karayannidis et al., 2004; Markl et al., 1999)
exploit such path-based surrogate keys, in order to achieve hierarchical clustering
of the fact table data.
In this chapter, we adopt a denormalized approach for the design of a dimension;
that is, we represent each dimension with only one table. The hierarchical attributes
(h1, h2, ??¦,hm), the feature attributes (f1, f2, ??¦, fk), as well as the hierarchical surrogate
key hsk of the dimension are stored in a unique dimension table. However,
the presented methods are fully applicable to normalized schemata (i.e., snowflaked
schemata) as well, with the only difference that extra joins between the several dimension
tables (corresponding to separate hierarchy levels) must be included in the
plan. In addition, we assume a special physical organization for the fact table. The
fact table is stored hierarchically clustered in a multidimensional data structure such
as the CUBE File (Karayannidis et al., 2004) or the UB-tree (Markl et al., 1999).
The index attributes of these structures are the h-surrogates.
Pages:
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285