There are a variety of
further physical design alternatives available for OLAP, for example, materialised
views (Gupta & Mumick, 1995) or data cubes. However, such can be seen somewhat
orthogonal to this discussion. The fundamental problems addressed here do
not change. Of course, combining these techniques is natural and will lead to even
better performance. But the general statements about the developed scheduling and
routing techniques would not essentially differ.
Finally, researchers have proposed data compression schemes and approximative
query-evaluation techniques (e.g., Chakrabarti, Garofalakis, Rastogi, & Shim, 2000),
including techniques that allow to trade result quality for query-answering time.
It should be noted that they are also orthogonal to our current concern, although
they complement each other very well: different cluster nodes could hold different
compressed versions of the database. The coordination middleware could then
take into account that more sophisticated compression schemes typically induce
higher maintenance costs. However, those combinations are beyond the scope of
this chapter. So in the following we assume the physical design on all cluster nodes
to be identical.
Query. Routing
For the following section, we concentrate on a query-only environment, and to support
the join-intensive OLAP queries, we assume full replication as physical design
in a cluster of databases.
Pages:
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450