For these results we consider TPC-H query set divided
into groups (Figure 12) according to sizes of replicated relations accessed and processed
by the joins. Group G1 accesses only a partitioned relation (LI or PS). The
other groups include references to replicated relations with sizes in the following
intervals:* Small: (0, 500MB); Medium: (500 MB, 5 GB); Large (5 GB, ???).
The speedup intervals (T lines) of Figure 13 are the range of speedup values considering
all queries in a group. From the results we can see that the larger the replicated
relations, the smallest the PRS speedup (G2, G3, and G4), with a large penalty for
processing heavy ReplicaJ joins. WBP achieved a near-to-linear speedup for all
queries, while PRS revealed very low speedup for most queries.
Figure 10. RT vs. Mbps (100 nodes)
Figure 11. Simulation vs. real execution (25 GB, 1Gbps)
0
00
000
0000
0 00 000
Mbps
response t me (secs)
PRS
PFRD-H
WBP
WBP+JB
0
200
400
600
800
0 0 00 200
n?? of nodes
responset t me (secs)
WBP-real
PRS-real
WBP-simula
PRS-simula
222 Furtado
Copyright ?© 2007, Idea Group Inc. Copying or distributing in print or electronic forms without written permission of
Idea Group Inc. is prohibited.
Replication for Nonstop
Availability on NPDW
In this section we discuss briefly alternative replication choices within the NPDW
for always-on, always efficient processing and allowing multiple nodes off-line simultaneously
for data loading, maintenance, and other DBA functionality.
Pages:
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422