SEARCH
0-9 A B C D E F G H I J K L M N O P Q R S T U V W X Y Z
Prev | Current Page 401 | Next

Robert Wrembel and Christian Koncilia

"Data Warehouses and Olap: Concepts, Architectures and Solutions"

g., every relation with
Figure 6. Node contents for PRS and PFRD-H partitioning strategies: (a) PRS
partitioning, and (b) PFRD-H partitioning
(a) (b)
2 6 Furtado
Copyright ?© 2007, Idea Group Inc. Copying or distributing in print or electronic forms without written permission of
Idea Group Inc. is prohibited.
less than 250 MB is to be replicated and those with less than 100 MB are
to be cached into memory for faster access). For our experiments we have
used this simple approach, but we describe a cost model and discuss the
search for optimal partitioning in the next section.
2. Facts: The objective is to find the hash-partitioning attribute that minimizes
repartitioning costs. A reasonable approximation to this objective
is to determine the most frequent equi-join attribute used by the relation.
To do this, the partitioning strategy looks at the frequency of access to
other partitioned relations and chooses the most frequent equi-join attribute
with those relations as the partitioning attribute. We have described
this process in the second section. A more complex approach involves
the search for optimal partitioning, as described in the next section.
By co-locating relation fragments that are frequent equi-join targets, this simple
strategy reduces significantly repartitioning requirements (we have determined experimentally
that WBP achieves an improvent of about 50% over straightforward
primary-key based partitioning (PK) when executing the query of Figure 5 under
the same conditions described later in the experiments).


Pages:
389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413