Information Technology Reference
In-Depth Information
1
0.95
0.9
0.85
00
01
10
11
0.8
0.75
0.7
0
500
1000
1500
2000
Time (hour)
Fig. 4.
Running time for some restrictions of data exchange
specific disks), while
b
2
indicates application of a limit on data migration (in the
case that
b
2
= 1, the migration capacity is limited to 10% of each disk space).
The figure shows the relative time when the running time in the case that all
the disks are active is equal to one. Here, the idleness threshold is fixed as 60
seconds in each configuration. In this simulation, we observed that our method
still reduced power even if one of the restrictions was introduced, which may
reduce the migration cost. In addition, under these parameter settings, the limit
on the exchange range (i.e.,
b
1
) was stronger than the limit on data migration
(i.e.,
b
2
), which would change depending on a given trace.
Accesses of Disks in Low-Power Mode.
Figure 5 shows the change in ratio
of the access of disks in low-power mode among all accesses in the same config-
uration as in the case of Figure 3. We observed that 6.8-28.7% of accesses were
of disks in low-power mode. This result means that 6.8-28.7% of accesses need
extra time for spinning-up. However, it should be emphasized that the data ex-
change realized a 9.5% reduction in accesses of disks in low-power mode, as seen
by comparing the “no-exchange” configuration and “30” configuration, which
had the same idle time but differed in terms of there being data exchange.
6 Experiments on an Implementation
We conducted an experiment on the current prototype implementation of our
proposed system to evaluate the applicability of our method to a real system. We
measured the response time and the cost of data migration in an environment
where the system workload was the same as that in the simulation.
Our prototype consisted of 50 PC servers (where one server was used for the
client and the other servers were used for Group B), each of which was equipped
