OLAP
queries are routed to one of the OLAP nodes depending on their data
freshness requirements. The lower their requirements are, the higher is
the probability that the coordinator finds a suitable OLAP node that (a)
is fresh enough and (b) is free to execute another query. If such a node
exists, the query will be immediately routed there and start executing.
Figure 1. Unified OLTP/OLAP cluster architecture
Clients
Coordination
Middleware
Database
Cluster
2 4 R?¶hm
Copyright ?© 2007, Idea Group Inc. Copying or distributing in print or electronic forms without written permission of
Idea Group Inc. is prohibited.
Otherwise, the query has to wait while the coordinator first updates an
OLAP node with previously logged updates to meet the reader??™s freshness
requirements, before it can route the query there.
There are a number of problems to solve with such a setting, for example, which
nodes to choose to evaluate an incoming OLAP query. But the central question is
how to efficiently propagate updates into the cluster without sacrificing correctness
or scalability. One solution is freshness-aware scheduling, which allows one
to seamlessly trade freshness of data for query performance.
System.Architecture
In the following, we give a more in-depth discussion of the system architecture as
developed in the PowerDB project at ETH Zurich (R?¶hm, 2002; Schek et al.
Pages:
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441