We will discuss a possible
solution to this update propagation problem in a later section of this chapter. The
central idea is to allow and consciously make use of multiple (consistent) versions
of the database at different cluster nodes.
The rationale behind hybrid designs is to combine the advantages of both data partitioning
and data replication. Consider for example the approach to partition only
the fact table while replicating the dimension tables (R?¶hm, 2000). With regard to
query evaluation, there is a distinction between queries that refer to the partitioned
relation and those that do not. In the first case, all cluster nodes process the original
query, and the coordination middleware computes the overall result. If the query
2 8 R?¶hm
Copyright ?© 2007, Idea Group Inc. Copying or distributing in print or electronic forms without written permission of
Idea Group Inc. is prohibited.
does not contain aggregation, the overall result is simply the disjoint union of the
intermediate results; computing the overall result in the other case is not difficult
either. If the query does not refer to the partitioned relation, their evaluation is as
with full replication.
While literature has proposed other schemes for physical organisation of databases
as a whole, for example, collocated joins (Baru et al., 1995) or multiattribute declustering
(Ghandeharizadeh, DeWitt, & Qureshi, 1992), one can see such techniques
as refinements of the basic alternatives described previously.
Pages:
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449