Download Print this page

Compaq 108164-003 - ProLiant - 800 White Paper page 25

Disk subsystem performance and scalability
Hide thumbs Also See for 108164-003 - ProLiant - 800:

Advertisement

Note: Within each test
comparison, the total disk
capacity remained constant.
ECG025.0997
W
P
HITE
APER
(cont.)
. .
. .
. .
Multiple Disk Controller Test Results
. .
. .
. .
In our multiple disk controller tests we found that concurrency coupled with adding disk
. .
controllers to an environment increases system performance. For instance, Figure 13
. .
. .
displays a comparison between two tests in a RAID 5 environment. First, we tested
. .
twelve 4GB drives (11 data, 1 parity) using one controller. Next using two controllers, we
. .
. .
equally split the number of drives (5 data, 1 parity) on each controller, totaling 10 data and
. .
2 parity drives. Even though we had two drives dedicated to parity in the two controller
. .
. .
environment, test results still show a significant benefit, yielding a 57% increase in
. .
system performance.
. .
. .
. .
. .
. .
. .
. .
. .
20,000,000
. .
. .
. .
18,000,000
. .
. .
16,000,000
. .
. .
14,000,000
. .
. .
. .
12,000,000
. .
. .
10,000,000
. .
. .
. .
8,000,000
. .
. .
6,000,000
. .
. .
4,000,000
. .
. .
. .
2,000,000
. .
. .
0
. .
. .
4
8
. .
. .
. .
. .
Figure 13: Disk Controller Scaling in a RAID 5 Environment.
. .
. .
The disk controller scaling tests configured with no fault tolerance or RAID 0, shown in
. .
Figure 14, receives only a minimum performance gain of 3% between one and two
. .
. .
controllers. Since we enhanced the disk subsystem and did not see any significant
. .
improvement, other factors must be limiting our throughput. Upon examination of the
. .
. .
other subsystems, we found the processors nearly saturated. To remove this bottleneck,
. .
we would have to use faster processors and rerun the test. Nevertheless, remember that
. .
. .
changing the disk subsystem enhances performance only if it is the bottleneck. In this
. .
case, the processors are the bottleneck and not the disk subsystem, so changing the disk
. .
. .
subsystem provides no benefit towards improving performance.
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
25
Disk Controller Scaling (RAID 5)
57%
16
20
24
28
32
Number of Clients
10+2 x 4GB, 2 Ctr
11+1 x 4GB, 1 Ctr
36
40
44
48
52
56
57

Advertisement

loading