On the
other hand, PMap retrieval cost remains the same or decreases with multi-attribute
queries, as the additional attributes limit the pstring search by changing
pfilterl and pfilterh values.
5. As the database size decreases, REBSI performance becomes better and the
relative savings of the PMap are reduced. This is intuitive since the number
of tuples directly impacts bit-vector length and the number of blocks to read
one bitmap decreases significantly. Savings for PMaps over REBSI are higher
for the larger database sizes and high cardinality attributes, essentially since
the REBSI performance deteriorates in these cases.
6. We identify two strategies to evaluate a disjunctive query, i.e., a query with
one or more OR operators. Strategy A splits a high level query into two or
more separate queries by rewriting it in disjunctive normal form and processing
each disjunct as a separate query, taking the union of the result set as the
final answer. These disjuncts are processed individually and the results for both
constitute the answer to the query. Strategy B processes a high level query as
a single query using multiple pfilters, without separating the disjuncts. We experiment
with both strategies for all the query sets having disjunctive queries
(VHCAQS, LCAQS, HSQS and MQS) and choose the best strategy in each
case in our performance studies.
Pages:
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384