.
..
(a)
(c)
(b)
(d)
248 R?¶hm
Copyright ?© 2007, Idea Group Inc. Copying or distributing in print or electronic forms without written permission of
Idea Group Inc. is prohibited.
Although the throughput improves only by less than 10% with cache approximation
routing as compared to FCFFS routing, the mean response time can be significantly
shortened. The best results achieved CAS-weighted routing, which offers almost
20% faster mean response times over the whole cluster size. All four investigated
routing strategies also showed a perfectly linear scalability over the cluster size.
For example, the throughput with a 24-node cluster is 24 times higher than on a
single node.
Performance.of.Freshness-Aware.Scheduling
In the following, we are interested in the performance characteristics of FAS. We
have used a dynamic workload of ten update streams concurrently executed with
twice as many querying clients as there are nodes in the cluster. We further varied
the mean freshness requested by read transactions from 0.6 up to 1. The results are
shown in Figure 5 (R?¶hm et al., 2002).
The presented middleware-based cluster architecture proved to be very scalable
and it also shows that freshness-aware scheduling effectively allows users to trade
freshness of data for faster query response time. We see that the slowdown of queries
by the concurrent update stream for FAS is around 10% up to 60% with regard
to mean response time as compared to the no-update case.
Pages:
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468