is prohibited.
and distributed databases. Replication strategies for shared-nothing systems range
from mirrored disk drives (Tandem, 1987) to chained declustering (Hsiao & De-
Witt, 1990a, b, 1991) or interleaved declustering (Teradata, 1985). Copeland and
Keller (1989) compare some of these high-availability media recovery techniques.
There are also recent works on replication (Coloun, Pacitti, & Valduriez, 2004; Lin,
Kemme, & Jimenez-Peris, 2005; Pacitti, ?–zsu, & Coulon, 2003), but the emphasis
is on transaction-related consistency issues. In general, most works focus generic
replication strategies for availability considering nonpartitioned relations and OLTP
workloads, while in this chapter we briefly discuss and evaluate replication on the
specific node-partitioned data warehouse context. An extended discussion on the
subject is available in Furtado (2005c).
Partitioning.and.Processing..............
over the NPDW
In a partitioning scheme, each relation can either be partitioned (divided into partitions
or fragments), copied in its entirety, or placed into a single node of a group
of nodes. We simplify the discussion by considering only one group (all nodes)
and homogeneity between nodes, in order to concentrate on the core partitioning
and processing issues. Generically, if a relation is large or very large, partitioning
is the choice that drives faster processing.
Pages:
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403