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 414 | Next

Robert Wrembel and Christian Koncilia

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


These results were obtained in a system with the characteristics: 50 GB TPC-H; 20
nodes, each with 866 MHz processor; 512 MB RAM). The alternatives compared
are: online??”every node is online; FPR??”fully partitioned replicas; PR(10)??”partitioned
replicas with two groups of 10 nodes; PR(5)??”partitioned replicas with 4
groups of 5 nodes. These results show the much larger penalty incurred by FR and
the excessive number of replicas required for FPR to allow 5 nodes off-line simultaneously.
PR(10) (partitioned replicas with two 10 element groups) is a good choice,
as it requires a single replica and obtains a good response time simultaneously.
Given these results, we conclude that replicas partitioned by groups are the most
advantageous alternative for NPDW if we consider both performance and flexibility
in allowing multiple nodes to be taken off-line simultaneously for maintenance
and loading reasons.
Figure 15. Response time and replicas when 5 out of 20 nodes are off-line (average
over TPC-H)
0
onl ne FPR PR( 0) PR( ) FR
nr of replicas
00:00
:
2 :0
4:
response t me (m n:sec)
N?? of Replicas
Response T me (m n:sec)
Efficient and Robust Node-Partitioned Data Warehouses 22
Copyright ?© 2007, Idea Group Inc. Copying or distributing in print or electronic forms without written permission
of Idea Group Inc. is prohibited.
Future.


Pages:
402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426