This preview shows page 1. Sign up to view the full content.
Unformatted text preview: r
numbers are better. Figure 1 compares the peak
number of virtual machines
running a Microsoft SQL
Server 2008 database
workload that each server
ran with acceptable
performance (all VM scores
higher than the baseline).
The Dell PowerEdge R710
ran 11 such simultaneous
VMs, while the HP ProLiant
DL385 G5 ran 9 such
simultaneous VMs. Thus, the
Dell PowerEdge R710
solution had a 22.2 percent
performance advantage over
the HP ProLiant DL385 G5 Microsoft SQL server 2008 Performance/watt results
Higher results are better
Normalized comparison 1.40
1.20 HP ProLiant DL385G5
server 1.00 Dell PowerEdge R710
0.00 Server solution. In addition, the Dell
PowerEdge R710 solution
had 3.0 percent more average
OPMs per VM than did the HP
ProLiant DL385 G5 solution.
As Figure 2 shows, the Dell
R710 solution delivered 21.4
percent more performance/
watt than did the HP DL385
G5 solution. We normalized
the results for each workload
to those of the server with
lower performance/watt. We
compute performance/watt by
dividing the benchmark’s
score by the average power
consumption, in watts, of the
solution during the period the
benchmark was delivering
peak performance. Figure 2: Performance/watt results for the solutions normalized to that of the HP ProLiant
DL385 G5 solution. Higher numbers are better. We installed and ran the DS2
client from VMs on a separate
ESX 3.5 host with four Intel X7460 2.66Ghz processors and 64 GB of RAM. We ran a single instance of the client
per VM, and each client VM targeted only one server VM. We assigned 2 virtual processors and 4 GB of RAM to
each client VM.
We defined the peak number of VMs per server as the maximum number of concurrent VMs under load where
each VM had at least the performance output in OPM as the baseline. Based on our analysis and experience with
DS2, we set the baseline at 2,600 OPM. Each VM on both the PowerEdge R710 and ProLiant DL385 G5 had to
score a minimum of 2,600 OPM to be valid. CPU utilization was near 100% on both servers when running at
peak VM levels. Workload
We conducted our testing using DVD Store Version 2, an open-source application with a back-end database
component, a front-end Web application layer, and a driver layer that operates as the middle tier and actually
executes the workload.
Because our goal was to isolate and test database server storage, we did not use the Web application layer.
Instead, we ran the driver application directly via its command-line interface.
DS2 models an online DVD store. Virtual customers log in; browse movies by actor, title, or category; and
purchase movies. The workload also creates new customers. Browsing movies involves select operations, some
of which use full-text search and some of which do not. The purchase, login, and new customer procedures
involve updates and inserts, as well as selects. The workload’s main reporting metric is orders per minute (OPM).
For more details about the DS2 tool, see http://www.delltechcenter.com/page/DVD...
View Full Document
- Fall '11