Professional Documents
Culture Documents
Table of contents
About this document
Executive summary
7
8
9
9
10
10
11
11
12
12
13
15
15
Validation methodology
Workload description
High availability
Unplanned downtime
16
16
16
16
Validation results
17
Conclusion
17
Appendix A
17
24
25
Figure 1: HP Client Virtualization with VMware View 600 User Logical Design
The architectural design is based on VMware View for Virtual Desktop provisioning and VMware vSphere for environment
virtualization. Storage components were implemented utilizing HP IO Accelerator card-based direct-attach solid state
storage, which is dedicated to hosting Virtual Desktops, and HP LeftHand VSA providing highly available storage for
persona. Each server was configured identically, thus the solution can scale from a single server to two and three
servers to ensure high availability.
The subsequent sections of this reference architecture document will provide full descriptions of the architecture and
test environment as well as performance metrics captured during test validation.
Executive summary
This stateless tiered storage reference architecture validation shows results for 200 Microsoft Windows 7 virtual
machines hosted on a single HP server. Using HP IO Accelerator and LeftHand Virtual SAN Appliance (VSA) powered by
SAN/iQ v9.5, VMware was able to achieve 200 Virtual Desktops per host with an average 94% CPU load per host. All
virtual machines were well below the accepted View Planner average response threshold of 2.5 seconds with a
maximum measured response time of 4.3 seconds. Consumed storage capacity grew modestly and averaged
approximately 350GB per host. HP LeftHand VSA was presented with enough hard disk space for persona, with a total
amount of 1.17TB of useable storage from the 4 enterprise SAS SFF HDD. Virtual storage bandwidth and IOPS utilization
was moderate to low. CPU was identified as the only performance limiting factor for scaling beyond 200 desktops per
host. At 200 users per desktop, costs for the VMware and HP components are as low as $420 per user plus the cost of
the client.
Additional highlights of the reference architecture IOPS and CPU usage are shown in Appendix A, and detailed
performance data can be found in the Validation results section.
PCoIP Optimization Controls: Deliver PC-over-IP (PCoIP) protocol efficiency and enable IT administrators to configure
bandwidth settings by use case, user or network requirements, and consume up to 75 percent less bandwidth
PCoIP Continuity Services: Deliver a seamless end-user experience regardless of network reliability by detecting
interruptions and automatically reconnecting the session
PCoIP Extension Services: Allows Windows Management Instrumentation (WMI)-based tools to collect more than 20
session statistics for monitoring, trending, and troubleshooting end-user support issues
View Media Services for 3D Graphics: Enable View desktops to run basic 3D applications such as Aero, Office 2010 or
those requiring OpenGL or DirectX without specialized graphics cards or client devices
View Media Services for Integrated Unified Communications: Integrate voice over IP (VoIP) and the View desktop
experience for the end user through an architecture that optimizes performance for both desktop and unified
communications
View Persona Management (View Premier Editions only): Dynamically associates a user persona with stateless
floating desktops. IT administrators can deploy easier-to-manage stateless floating desktops to more use cases
while enabling user personalization to persist between sessions
Because its based on VMware vSphere 5, VMware View leverages the latest functionality of the leading cloud
infrastructure platform for highly available, scalable, and reliable desktop services.
For additional details and features available in VMware View 5, see vmware.com
VMware View is orderable from HP at
http://h18006.www1.hp.com/products/servers/software/vmware/virtualization/vdi/index.html.
HP supports all pieces within the solution so there is never any finger pointing.
Architecture for VMware View found at hp.com/go/cv. Combined with VMwares storage accelerator (host caching) this
allows customers to achieve lower costs per user while still providing a compelling user experience and impressive
storage performance.
The stateless architecture provides linear scalability from a single host to hundreds of hosts, and allows a modular
approach for customers to scale out their virtual desktop infrastructure.
This design is intended to be a general guide that addresses common design questions for stateless desktops.
A single-user role type was chosen to demonstrate scalability with a typical use case. By virtualizing the application set,
a single pool type could be deployed within an environment with multiple user roles.
The abstraction of the applications from the base image allows each user to leverage the same base OS and multiple,
different applications thereby reducing the administrative overhead required to maintain different user roles in the
same VDI architecture.
Compared to HP ProLiant G7
Based on HP internal testing with a 50% increase in server performance (OLTP applications), a 10% reduction in per server energy consumption plus 7%
cooling reduction from Intelligent Series rack. Assume a conservative PUE of 2.0 translates 7% cooling to 3.5% reduced cooling power.
The math : 10% (< per server) + 3.5% (< from Intelligent Series rack) = 13.5%.
150% (performance) x 113.5% (servers/P&C envelope) = 1.5 x 1.135 = 1.70
3
HP call center data based on an analysis of incident records during Q411. For customers using HP Insight Remote Support (IRS), their hardware-related
problems were solved 66% faster compared to customers not using IRS and 95% of the time it was fixed on the first try.
Based on analysis of over 12,000 incident records per month across all regions for Q411 where an onsite visit or spare part was required.
Server/storage
TESTED Configuration
SFP+ Adapter
vSphere ESXi 5.0 Update 1 Build 623860
3 HP 365GB Multi Level Cell G2 PCIe ioDrive2 for ProLiant Servers
Note: The test systems were configured with larger 785GB
storage
Each host configured with 4 x HP 300GB 6G SAS 10K 2.5in SC ENT
HDD
Note: The HP ProLiant DL380p used Intel Xeon ES-2680 processors to facilitate what was deemed a compelling price/performance
mix. It is possible to upgrade to the Intel Xeon ES-2690 processor for enhanced performance.
Virtual Machine
Active Directory/DNS
2 vCPUs
2 GB memory
1 Ethernet card
Virtual HDD (1 x 45 GB)
Virtual Machine Disk (vmdk) only
2 vCPU
4 GB memory
1 Ethernet card
Virtual HDD (1 x 40 GB)
vmdk only
View Manager
2 vCPU
4 GB memory
1 Ethernet card
Virtual HDD (1 x 40 GB)
vmdk only
10
Virtual Machine
DHCP Server
2 vCPU
2 GB memory
1 Ethernet card
Virtual HDD (1 x 15 GB)
vmdk only
1 vCPU
1 GB memory
1 Ethernet card (10 GbE)
Virtual HDD for VSA installation on local datastore (1 x 7 GB)
Virtual HDD attached to VSA (1 x 820 GB)
vmdk only
HP IO Accelerator configuration
Each server was configured with single HP IO Accelerator. Given the stateless design, it is not necessary to have a
mirrored array of solid state disks.
If a host, drive, power supply or similar component fails, the high-level design provides the redundancy necessary to
provide a user with a new desktop as detailed in the High availability section of this reference architecture document.
Capacity is the center point of the design, as the ability to use the local HP IO Accelerator for various components of the
desktop is only limited by todays technology.
To make this architecture cost effective, a capacity of 365GB is more than sufficient to support a properly optimized
Windows 7 desktop at the quantities listed in this document.
User data, the parent base image and the .vswps associated with the individual desktops were on shared storage. The
local HP IO Accelerator contained only the replica base image and linked clones.
To ensure realistic results with a new solid-state drive, special attention is required when being used for benchmarking.
A new solid-state storage device is faster with writes when new than it will be after heavy use. This is especially
prominent with MLC-based solid-state. While this is a well understood issue that is easy to address, without
preconditioning the results from the local IO Accelerator would be misleading. Preconditioning may be achieved by
running two or more full test iterations prior to finalizing test results. HPs standard test practice is to fill the accelerator
completely, running two test runs and then running a final test to measure expected results.
11
Access infrastructure
The physical networking was implemented with a redundant network core of full 10Gbit Ethernet. The network core also
load balances incoming requests across VMware View Connection Servers, where user requests were routed to the
appropriate building block for each virtual desktop session.
Realistically, it would be possible to provide networking at 10Gbit speeds for 3 servers with a pair of switches; however
this reference architecture used core networking as it would be more typical in a data center.
Description
VLAN CONFIGURATION
VLAN ID
Description
12
13
14
Windows 7 32-bit
RAM
1GB
Number of vCPUs
24GB
15
Validation methodology
When validating the VMware View for HP Client Virtualization design, it is important to simulate a real world
environment as closely as possible. For this validation, each component was built and validated for the virtual
infrastructure necessary in a 200-user server building block using a simulated workload. The networking, virtual storage
appliance, and core common infrastructure components were also implemented and tested for the access infrastructure
necessary to support a three-server solution.
Testing was conducted in two phases. In the first phase, desktop pools were created provisioning the virtual machines to
a single server. Each pool was created manually, just as it typically would be created in a normal environment. Testing
was performed to find a tested limit of the single server, which for the server described in this document was 200
desktops.
The second phase of the validation included session establishment, logon, and execution of a workload across the
cluster. Client access devices established individual VMware View sessions, and connected to an assigned View desktop
using the VMware View Client. Once a session was established, a workload was run to simulate typical user activity.
Each session worked in the environment as a standard user throughout the test, during which overall system statistics
were collected from several components of the architecture. The following sections explain in more detail how each
layer was implemented and used as part of the validation and how the workload was implemented.
Workload description
Each virtual machine was equipped to run VMwares View Planner workload that simulates typical user behavior. View
Planner uses an application set comprised of Microsoft Office 2007 applications. The workload has a set of randomly
executed functions that perform operations on a variety of applications. Several other factors can be implemented to
increase the load or adjust the user behavior, such as the number of words per minute that are typed and the delay
between applications being launched.
The workload configuration used for this validation included Office 2007: Word, Excel, Outlook and PowerPoint.
During the execution of the workload, multiple applications were opened at the same time and windows were minimized
and maximized as the workload progressed, randomly switching between each application. Individual application
operations that were randomly performed included:
Microsoft Word 2007: Open/minimize/close, write random words/numbers, and save modifications
Microsoft Excel 2007: Open/minimize/close, write random numbers, insert/delete columns/rows, copy/paste
formulas, and save modifications
High availability
The VMware View for HP Client Virtualization design has different requirements for high availability as compared to a
standard desktop virtualization deployment. In the past, the design would be based on the redundancy provided by a
shared storage device, and would utilize features within VMware vSphere to provide High Availability (HA), Distributed
Resource Scheduling (DRS), and live migration with vMotion.
In this design, VMware HA is not necessary, as the failure of a host will be seen by the VMware View Connection Servers.
In case of server and/or storage failure, the broker will simply allocate a new desktop for the user after successful
authentication. A highly available implementation of a Microsoft Distributed File System (DFS) share enables always-on
View Persona provisioning. HP LeftHand VSA will sustain a single host failure with an HP LeftHand FOM properly
configured when an even number of VSA nodes are deployed or when odd node counts of 3 or greater are used. VMware
View Connection Servers can address both planned and unplanned downtime at the broker level, which was tested
extensively as a part of this reference architecture.
Unplanned downtime
A standard host-level outage is easily mitigated by careful pool planning. By provisioning a surplus number of virtual
desktops across the cluster, equal to the number of failed virtual desktops on a given host, the failure of a host is
contained. In the design, 200 desktops will be created on each host; however, only 130 would be allocated to users by
the broker unless there was a host/storage failure.
16
As the Connection Server will automatically sense that the desktops are no longer available after 30 seconds, if a user
that was on the failed server attempts to log back in, the user will receive a new desktop from another host. This
effectively provides HA, but at the Connection Server layer.
Validation results
This section details the results and observations that were concluded during the validation of stateless tiered storage of
desktop virtualization with VMware View 5.
All validation included several test iterations to ensure data consistency after a ramp-up period; for clarity multiple
iterations are not shown.
Testing was performed with an optimized Windows 7 image following VMware best practices. All data includes the initial
period of massive user logons ramping up the test to demonstrate that even when pushed heavily the environment can
provide an acceptable range of performance.
The storage pieces of the architecture proved capable of providing more than enough throughput and IOPS capacity. The
HP IO Accelerator latency averaged around 1.8ms and LeftHand VSA averaged 8ms. HP IO Accelerator IOPS was not even
challenged at an average of 900 IOPS, peaking at 1300 IOPS. LeftHand VSA had an even lower load of 11 IOPS on
average and peaking at 250 IOPS. For more details please see the figures in Appendix A.
With respect to CPU, the system used approximately 94% of CPU at boot up. However, once the systems reached a
steady state, utilization dropped to only 8%. For more details please see figure A-8 in Appendix A. Even at full CPU load
desktops were responsive and with an average application response time of 1.2 seconds (well below the accepted View
Planner average threshold of 2.5 seconds) and max value barely reaching 4.3 seconds. For more details see figure A-1 in
Appendix A.
This architecture shows that even in an HA scenario with three servers and an HA friendly 120-130 desktops per host
configuration (60% server load), an architecture can sustain unplanned single node failure and still provide enough
computing capacity to host all desktops on the remaining 2 servers.
The initial ramp-up period shows higher latency, but as mentioned earlier this was included to show how well the
infrastructure deals with a massive influx of all users logging in all at once.
Capacity of the HP IO Accelerator was tracked closely to ensure the environment would have no issue supporting the
necessary user counts. Even after multiple test runs, the linked clones grew only modestly. The architecture should
allow for a weekly refresh period.
Conclusion
The VMware View 5 architecture for HP Client Virtualization discussed in this reference document significantly reduces
the hardware infrastructure costs of desktop virtualization environments. The architecture provides a stateless desktop
virtualization environment designed with VMware View 5 tiered storage that can scale based on demand while providing
the lowest cost per desktop in the industry. In summary, this design provides linear scalability across both compute and
storage.
Appendix A
The View Planner workload was configured with different standard applications like Word, Excel, PowerPoint and
Outlook. The graphs in this section represent the response times for different operations of the applications as well as
overall performance as discussed throughout the document.
17
Figure A-1: Maximum Application Response Time of 4.3 Seconds, even at 200 Virtual Desktop workload
18
19
20
Figure A-6: Maximum 2.4ms Read Latency to the local HP IO Accelerator datastore
21
Figure A-7: Host Memory usage was at a stable 235GB throughout the test
22
23
Qty
Part Number
Description
653200-B21
662228-L21
662228-B21
48
647901-B21
12
652564-B21
684210-B21
631681-B21
656363-B21
BC390AAE
673642-B21
TA688AB
JG296A
HP 5920AF-24XG Switch
BD691AAE
24
Get connected
hp.com/go/getconnected
Current HP driver, support, and security alerts
delivered directly to your desktop
Copyright 2012 Hewlett-Packard Development Company, L.P. The information contained herein is subject to change without notice. The only
warranties for HP products and services are set forth in the express warranty statements accompanying such products and services. Nothing herein
should be construed as constituting an additional warranty. HP shall not be liable for technical or editorial errors or omiss ions contained herein.
Microsoft and Windows are U.S. registered trademarks of Microsoft Corporation. Intel and Xeon are trademarks of Intel Corporation in the U.S. and
other countries.
4AA4-3070ENW, Created August 2012
25