Specifications

A Principled Technologies test report 13
Cisco UCS B200 M3 Blade Server:
Uncompromised virtual desktop performance
APPENDIX B – HOW WE TESTED
To determine the number of virtual desktops the server could support, we ran incremental tests increasing the
virtual desktop load until Dynamic VSI Max was reached. The minimum number of sessions required to archive Dynamic
VSI max was 193. When testing 193 virtual desktops, we recorded a Dynamic VSI max of 186. At 192 sessions, the
benchmark did not achieve Dynamic VSI Max. At the end of the test, the processors on the Cisco UCS B200 M3 Blade
Server were nearly saturated. We confirmed that no other factors such as storage bottlenecks or memory constraints
were contributing to a loss of performance or decreased response time.
Figure 8 illustrates our test environment: one Cisco UCS 5108 Blade Chassis with one Cisco UCS B200 M3 Blade
Server and one Cisco UCS B200 M2 blade server. The Cisco UCS B200 M2 blade server with VMware vSphere 5 hosted all
VMware View 5 and Login VSI Infrastructure VMs and the Cisco B200 M3 blade server with VMware vSphere 5 hosted all
the VMware View 5 virtual desktops. We connected the Cisco UCS 5108 Blade Chassis to redundant pair of Cisco UCS
6248UP Fabric Interconnects. We connected the Fabric Interconnects to a Cisco Nexus™ 5010 switch. We deployed the
two Cisco blade servers via the Cisco UCS Manager. Using Cisco Service Profiles, we assigned a base firmware level of
2.0(2a) for all server components. For the Cisco B200 M3, we assigned two redundant 20Gb vNICs and two redundant
4Gb Fibre Channel HBAs. For the Cisco B200 M2, we assigned two redundant 10Gb vNICs and two redundant 4Gb Fibre
Channel HBAs. We hosted all VM storage on an EMC VNX 5500. We set up our VMware View virtual desktops via a
linked clone pool. Our master image, a Microsoft Windows 7 x 86 Enterprise VM, had one vCPU and 1.5 GB of reserved
memory. We used the VMware View script commands.bat found at http://www.vmware.com/files/pdf/VMware-View-
OptimizationGuideWindows7-EN.pdf.
We disabled the build to lossless feature and set the maximum frames per second to 15 as discussed at
http://www.vmware.com/files/pdf/view/VMware-View-5-PCoIP-Network-Optimization-Guide.pdf
All Login VSI launchers were virtual machines. To ensure audio rendering, we installed Virtual audio cable 4.12
on each. All desktop session were at a resolution of 1024x768, 32 bit color.
Note that, for its UCS blade servers, Cisco recommends a stateless boot from SAN configuration to ensure
portability. However, for simplicity we installed our vSphere operating system on local disks because it did not affect the
performance testing.
Figure 9 illustrates our logical network layout. We created a vSwitch on each vSphere server and created a single
port group, tagged as vlan100. We connected all virtual desktops, Login VSI launchers, and VMware View infrastructure
to these two vSwitches.