QoS-Oriented Grid-Enabled Data Warehouses - Data Warehousing Design and Advanced Engineering Applications - page 165

Figure 7. SLO-achievement rate

number of data replicas. But as the BC method
places too much replicas at such site, the scheduler
assigns too much queries for site 6 (what puts it
into a too high load situation and decreases per-
formance, leading to a lower SLO-achievement
rate) or schedules queries to other sites that are
near site 6 but that do not have the same data
replicas (in such case, data movement is done
during query execution, which also decreases the
system performance).
On the other hand, the QoS-aware placed
high number of replicas at sites 3-5 than the other

methods. Such sites are of medium powerful ones
and are relatively near of sites 1, 2 and 6 (the
three sites with the highest number of submitted
queries). This way, many of the queries submit-
ted at sites 1, 2 and 6 can be executed at sites 3-5
with a good performance (no database copy is
done during execution, only results - which are
relatively small in size - are transferred through
sites during query execution). In contrast, the DLL
strategy placed many replicas on not so powerful
sites (like 10 and 7). This happened as the DLL
strategy evaluated that such sites had a small size

Figure 8. Measured throughput