Technical white paper

HP Reference Architecture for Client

Virtualization
Client Virtualization from HP with Citrix XenDesktop and Microsoft Windows Server
2008 R2


Table of contents
Executive summary 3
Client virtualization and HP 4
Purpose of this document 4
Reference Architecture components 5
HP BladeSystem 5
HP ProLiant Servers 5
HP Storage 6
User data storage 8
HP Virtual Connect Flex-10 9
HP 12500 Series Switches 10
HP Thin Clients 11
Citrix XenDesktop 11
The best choice for desktop
virtualization 11
Provisioning in XenDesktop 12
Microsoft Windows Server 2008 R2 13
System Center 2012 Virtual Machine
Manager 13
Reference Architecture overview 14
Core design 14
Understanding the VDI Client
Virtualization Reference Architecture 14
Persistent and non-persistent VMs 18
The Reference Architecture 19
Testing methodology 19
Test environment 19
Real world vs. lab 23
The non-persistent model 24
Management VM configurations 26
The persistent model 28



HP, Citrix, and Microsoft better
together 31
Summary 32
Appendix A 32
For more information 34
Call to action 34






3
Executive summary
Businesses are looking to enable flexible workstyles for their users with the ability to work anytime, anywhere, from any
device while increasing IT efficiency through cloud services. Client virtualization is a hot topic because it addresses these
two pain points.
However, client virtualization can open a Pandoras Box of terms, ideas, and questions. Client virtualization can be as
broad as providing access to a remote desktop to an end user to as specific as delivering a non-persistent streamed
desktop, or virtual desktop infrastructure (VDI).
A hosted desktop solution, formerly known as server based computing, is straight forward and has been around for
many years in the familiar Citrix XenApp/Metaframe products. In this model all users share the same desktop interface
hosted on a server on the back-end, with no opportunity for user customization. Updates need only be applied to the
server to affect all of the users. This approach works very well for entry level users running only a few apps and no
requirement to modify or customize their environment. For users requiring administrator rights, specific unique
applications or isolation from other users for security a virtual desktop infrastructure can be utilized.
For a virtual desktop, persistent and non-persistent users need to be considered. A persistent user maintains changes to
their desktop between logins. A non-persistent user gets a fresh new desktop from a base image every time they log in.
The primary difference between a hosted desktop user and a non-persistent user is user isolation in a dedicated virtual
machine. What one user does in the hosted desktop environment can affect all other users in that hosted environment,
as all users are sharing the memory and CPU of the server.
In a non-persistent VDI environment all users share the same base image file, with any changes or modifications they
make stored in a differential file, normally a few GB in size, and each user is allocated virtual CPU(s) and memory. When
the user logs off, the differential files are deleted and re-created for the VM. If they exceed the capabilities of the
assigned processing and memory it only affects that users performance. Updates are applied to the base image and all
users will get the updates.
Like a non-persistent user, a persistent VDI user is isolated from other users on the same server, but the difference is
the user customizations are maintained between logins in an assigned virtual hard disk (vHD), one file per user. This
model generates many vHD files to manage. Another approach for persistent users is similar to non-persistent users,
but the differential files are maintained across logins so any changes the user makes will be stored in the differential
files. However, if the base master image is ever changed one of two things will happen depending on the configuration.
The first is the differential files are marked invalid, and the next time the user logs in they are deleted and re-created,
and the user loses any customization they did. The second is the changes in the base image will be copied to all of the
differential files, increasing the size if the differential files. Over time these differential files could grow to be quite large.
In many cases instituting a client virtualization solution is not a single solution, but a hybrid strategy. Companies that
have tried to use hosted desktops for all users have found that not all users fit into the entry level user, nor do they all
use the same applications.
Persistent desktops allow for isolation of users, and allow different user to have different applications. However,
companies that have simply converted their existing desktop images into persistent desktops may lead to large and
difficult to manage images, excessive storage consumption and may lead to a more costly model to support than
dedicated desktops.
For VDI, the non-persistent user is the most desired model. Benefits of a non-persistent user model include:
A single or only a few images to maintain.
No one user can affect other users.
The amount of required storage can be reduced.
However, and there always seems to be a however, users cannot customize their environments unless user
virtualization is implemented using Citrix User Profile Manager or third party applications to manage the user
environment. User virtualization allows for each user to customize their environment, and maintain that customization
across logins.
Citrix also provides the FlexCast model, allowing for multiple ways to deliver the desktop to the end user, from using the
hosted desktop capabilities of Citrix XenApp, to creating a virtual desktop that is streamed to an end user device using
XenDesktop and Provisioning Server. And now with XenDesktop 5.6, Citrix has added personal vDisks, PvD. One of the
downsides to non-persistent VMs even with user virtualization is allowing a user to install customized or specific
software that only they have a need to access. The PvD from Citrix allows a non-persistent user to become a persistent



4
user, but still use the same base image. These users still boot from the master image maintained by IT, but have the
capability to customize their environment, even install software, and have those changes remain across logins.
This document will focus on client virtualization using VDI technology with XenDesktop, and the
considerations/implications of persistent vs. non-persistent users.
HP has partnered with Citrix to leverage the converged capabilities of HP servers, storage and networking to develop a
VDI client virtualization (CV) Reference Architecture optimized to take advantage of Citrix FlexCast. Key differentiators of
this reference architecture include:
Simplicity: Whether hosting a 100-user proof-of-concept or a 20,000 user production implementation, the building
block remains the same. HP BladeSystem allows for the use of high scaling, distributed 2-socket servers in VDI
implementations while maintaining a minimal amount of infrastructure to be managed. The result is a highly available,
power efficient, simple to manage infrastructure that sacrifices nothing while delivering optimized costs, converged
storage and networking, and simple scalability. This Reference Architecture provides validated configurations for a
single enclosure containing the necessary compute, storage and network required to support 1,690 users and
leveraging Virtual Connect across racks to support over 6,500 users. Each enclosure has all the servers, storage and
network interconnects required to run XenDesktop.
Network flexibility with reduced costs: Leveraging HP Virtual Connect technology to deliver network flexibility and
performance while reducing costs is accomplished by migrating as much of the infrastructure and compute resources as
possible to within the Virtual Connect domain.
Optimize storage price/performance: This RA takes into consideration the unique characteristics of Citrix XenDesktops
Provisioning Server and personal vDisks and provides optimal configurations using the HP Virtual SAN Appliance (VSA)
that maximize the usage of lower cost Direct Attached Storage and minimize the use of higher-cost SANs.
Rich user experience: HP has a line of Citrix Ready zero and thin-clients optimized to take advantage of Citrix HDX
technologies to deliver a consistent, rich end-user experience delivering an end-to-end solution
Target audience: This document is intended for IT decision makers as well as architects and implementation personnel
who want to better understand HPs approach to client virtualization. The reader should have VDI knowledge and
understanding, understanding of XenDesktop, Microsoft Windows Server 2008 R2 and Microsoft System Center Virtual
Machine Manager (VMM) 2012, and understands sizing/characterization concepts.
This and other documents pertaining to client virtualization with HP may be found at hp.com/go/cv.
This white paper describes testing performed in March of 2012.
Client virtualization and HP
HP has been involved with client virtualization for many years, from HPs working with Citrix for hosted desktops with
XenApp
1
, previously known as Presentation Server, creating the Consolidated Client Infrastructure using PC based
blades in the early 2000s, to HP Workstation Blades
2
allowing 1-to-1 connections delivering hardware accelerated 3D
graphics to remote desktops, to working closely with Citrix and their development of XenDesktop as a product.
Purpose of this document
This paper focuses on HPs recommended approach to virtual desktops in client virtualization with Citrix XenDesktop
5.6, Citrix Provisioning Server 6.0, Microsoft Windows Server 2008 R2, and VMM 2012 using HPs blades and storage to
create a reference architecture. The document will discuss HPs testing methodology, how testing and characterization
was done, and highlight best practices in setting up a reference architecture. This is not a step-by-step instruction guide
to build a reference architecture. This document will not give installation guidance, or show how to install the software
discussed. It will highlight considerations and best practices to help ensure a successful VDI implementation.
This document will focus on the hardware and configurations that HP considers to be the sweet spot around VDI.

1
HP has been working with Citrix around hosted desktop sizing and configuration. Information can be found at
http://h20338.www2.hp.com/enterprise/cache/3916-0-0-0-121.html.

2
For more on HP ProLiant WS460c Workstation blade, go to hp.com/go/workstationblades



5
Reference Architecture components
HP BladeSystem
Drive business innovation and eliminate server sprawl with HP BladeSystem, the industrys only Converged
Infrastructure architected for any workload from client to cloud. HP BladeSystem is engineered to maximize every hour,
watt, and dollar, saving up to 56% total cost of ownership over traditional infrastructures.
With HP BladeSystem, it is possible to create a change ready, power efficient, network optimized, simple to manage and
high performance infrastructure on which to build and scale your VDI implementation. An HP BladeSystem c7000
enclosure populated with HP ProLiant BL460c Gen8 blades is shown in Figure 1.
Figure 1: The HP BladeSystem c7000 enclosure (pictured with HP ProLiant BL460c Gen8 server blades)

Benefits for VDI: HP BladeSystem allows for the use of high scaling, distributed 2-socket servers in VDI
implementations while maintaining a minimal amount of infrastructure to be managed. The result is a highly available,
power efficient, simple to manage infrastructure that sacrifices nothing while delivering optimized costs, converged
storage and networking, and simple scalability.
HP ProLiant Servers
Choosing a server for VDI involves selecting a server that is the right mix of performance, price and power efficiency with
the most optimal management. HPs experience during test and in production has been that 2 socket servers are the
ideal platform for VDI. With the potential of over 100 VMs running on one platform, 2 socket systems offer better
memory performance and thus better scaling models than larger systems. HP BladeSystem reduces costs and simplifies
management through shared infrastructure.



6
HP ProLiant BL460c Gen8
The HP ProLiant BL460c Gen8 server blade offers the ideal balance of performance, scalability and expandability for any
workload, making it the standard for dense data center computing.
Figure 2. HP ProLiant BL460c Gen8 server blade

The HP ProLiant BL460c Gen8 offers Intel Xeon E5-2600 series processors, Flex adapters supporting Flex-10
3
with
Ethernet, iSCSI and even FCoE traffic, up to 512GB of DDR3 memory, two hot plug drive bays with an embedded HP
Smart Array controller as well as industry leading HP Integrated Lights-Out 4 (iLO 4) management capabilities.
Benefits for VDI: The BL460c Gen8 is an ideal platform for all VDI user types. The BL460c offers not only high user
counts per host, but also incredible density which is far more important in optimizing power and management efficiency
as well as reducing infrastructure. It also makes no compromises with memory speed which plays an important role over
capacity when achieving maximum performance.
HP Storage
Storage is as important as the servers in your client virtualization deployment. The storage infrastructure is directly in
the critical path for hundreds or thousands of your end users and this introduces a number of unique challenges with
respect to delivering consistently high performance, minimizing management and deployment costs, and scalability and
efficient capacity utilization. For this reason, the storage platform you choose is critical to making the economics of
client virtualization work while maintaining high quality of service levels to satisfy user requirements. HP provides
storage solutions to meet small to large client virtualization deployments with a range of SAN, NAS and DAS solutions.
HP D2000 Disk Enclosures
The D2000 disk enclosures deploy with the next generation 6Gb SAS for a flexible tiered external storage system. The
6Gb SAS enclosures - Large Form Factor (LFF) D2600 with 12 drive bays and Small Form Factor (SFF) D2700 with 25 drive
bays - offer modular solutions to simplify capacity expansion of HP ProLiant server environments to external storage
without having to make the full move to SAN or NAS. This allows you to buy what is needed today and purchase
additional capacity as data storage needs grow.
The HP D2700 w/25 300GB 6G SAS 10K SFF Dual Port HDD 7.5TB bundle is utilized to supply direct attached storage to
the blades through 6Gb SAS switches. Expansion can be accomplished by adding additional D2700 disk enclosures and
using the SAS management console integrated into the switch to assign the drives to blades.

3
More information about HP Flex technology can be found at: http://h20000.www2.hp.com/bc/docs/support/SupportManual/c01608922/c01608922.pdf



7
Figure 3: The HP D2700 Disk Enclosure Front and Rear view
UID
HP
StorageWorks
D2700
SAS
6G DP 15k
146 G
B
SAS
6G DP 15k
146 G
B
SAS
6G DP 15k
146 G
B
SAS
6G DP 15k
146 G
B
SAS
6G DP 15k
146 G
B
SAS
6G DP 15k
146 G
B
SAS
6G DP 15k
146 G
B
SAS
6G DP 15k
146 G
B
SAS
6G DP 15k
146 G
B
SAS
6G DP 15k
146 G
B
SAS
6G DP 15k
146 G
B
SAS
6G DP 15k
146 G
B
SAS
6G DP 15k
146 G
B
SAS
6G DP 15k
146 G
B
SAS
6G DP 15k
146 G
B
SAS
6G DP 15k
146 G
B
SAS
6G DP 15k
146 G
B
SAS
6G DP 15k
146 G
B
SAS
6G DP 15k
146 G
B
SAS
6G DP 15k
146 G
B
SAS
6G DP 15k
146 G
B
SAS
6G DP 15k
146 G
B
SAS
6G DP 15k
146 G
B
SAS
6G DP 15k
146 G
B
SAS
6G DP 15k
146 G
B

UID
PS 1
PS 2
2 1
I/O B
I/O A
Mfg
P2 P1
UID
Mfg
P2 P1
UID
Attention
This module fits 2.5" small
form factor drives only.
Attention
This module fits 2.5" small
form factor drives only.

Benefits for VDI: The D2700 7.5 TB bundle allows for ease of growth by adding additional disk enclosures to support
additional blades. Configuration for each disk enclosure is handled using the SAS management console and the HP ACU,
array controller utility.
HP LeftHand Storage Solutions ideal storage for client virtualization
HP LeftHand Storage solutions are highly optimized for virtual server and desktop environments, with a storage
architecture that enables non-disruptive linear scaling of capacity and performance so you can grow your client
virtualization solution as you need, when you need. With a wide range of storage solution deployment options from a
virtual SAN appliance to rack and blade systems HP LeftHand Storage provides the optimal mix of storage affordability
with the right set of features and scalability to meet your client virtualization storage needs, no matter how small or
large. HP LeftHand Storage delivers:
Storage Efficiency to Maximize Your client virtualization Investment:
Thin provisioning raises storage utilization and efficiencies
Clustering, density, and storage tiers optimize dollars per I/O
High Performance to Enable Worker Productivity:
Clustered, virtualized architecture delivers maximum disk utilization
Dense spindle count, high-speed storage paths, optimized use of SSDs
High Availability to Support Critical End User Applications:
No single-point-of-failure storage architecture through Network RAID
Non-disruptive software upgrades
Wide Ranging Scalability to Grow When You Need:
Linear, non-disruptive capacity and performance scaling
Protect your capital investment, start small and grow large
For more information, see hp.com/go/lefthand.
HP LeftHand P4800 Storage Solutions for HP BladeSystem
The HP LeftHand P4800 Storage Solution delivers storage that is fine-tuned for client virtualization with HP
BladeSystem. It is based on a tight integration with HP BladeSystem and offers a highly dense storage solution that
dynamically and linearly scales as the infrastructure expands. Being converged into the HP BladeSystem architecture
enables simplified administration, improved security, flexibility, and enhanced performance through network
convergence of the virtual server and storage fabric using 10Gig Ethernet and Flex-10 Virtual Connect.
Convergence with the HP BladeSystem architecture also provides a significant density and footprint advantage further
saving costs. HP P4800 Storage for BladeSystem is designed for departmental or enterprise environments with greater
than 500 users.



8
Scaling from 2 nodes and 70 disks (see Figure 4) to 8 nodes with 280 disks in a single BladeSystem c7000 enclosure with
associated storage drawers, or to 16 nodes with 560 disks across multiple BladeSystem c7000 enclosures and storage
drawers ensures I/O performance is outstanding for the large scale VDI deployments that HP LeftHand P4800 addresses.
The ability to make private storage networks where storage traffic remains within the Virtual Connect domain means
one administrator can manage the entire infrastructure stack without the need for specialized storage skill sets or in
depth network administrator involvement.
Figure 4: The HP LeftHand P4800 Storage Solution for HP BladeSystem

Whether you are looking to place persistent VMs or non-persistent VMs, the HP P4800 offers up the mix of capacity with
thin provisioning and performance required for VDI.
Benefits for VDI: The convergence of not only storage but also the transport layer into a single, simple to manage
infrastructure means intensive multi-team and multi-skill set involvement is a thing of the past. This leads to more rapid
deployment and faster response to business-driven change requests. The fact that the fabric is integrated across high
speed electrical links leads to performance that is both impressive and scalable, so your users are always productive.
HP LeftHand Virtual SAN Appliance Software
HP LeftHand Virtual SAN Appliance is a VM running the SAN/iQ software allowing local or direct attached storage to be
used to create volumes and clusters to be shared between servers.
Benefits for VDI: Utilizing HP LeftHand Virtual SAN Appliance Software (VSA) for selective use as a storage layer allows
for migration of infrastructure VMs, creation of volumes for clustering between hosts and SQL servers, and storage for
provisioning servers. VSA is even managed from the same console as the other HP LeftHand storage tiers and scales
simply to meet demand.
User data storage
Data centers today rely on a variety of storage solutions to support different networks, applications and users.
Flexibility and scalability are key to meeting these various and often changing requirements. HP X3800 Network Storage
Systems meet the requirements of any size data center. For example, the HP X3800 Network Storage Gateway, a
network-attached storage (NAS) appliance, with Windows Storage Server 2008 R2, provides large data centers the
performance, flexibility and scalability they need to support tens of thousands of users. HP recommends keeping user
data on external storage outside of the VDI environment. Figure 5 shows the HP X3800.



9
Figure 5: HP X3800 Network Storage Gateway

Benefits for VDI: The HP X3800 Network Storage Gateway integrates into existing Microsoft Windows management
processes and brings user data into data center backup plans while keeping management simple.
HP Virtual Connect Flex-10
HP VDI and Flex-10 technology
HP Virtual Connect Flex-10 technology creates dynamically scalable internal network architectures for VDI deployment.
Multiple c7000 enclosures make up the core building blocks of the physical infrastructure. Each enclosure contains two
Virtual Connect (VC) Flex-10 interconnect modules. Each module connects to two dual port 10Gb Flex-10 or FlexFabric
adapters in each server (Gen8 BladeSystem servers have embedded FlexFabric adapters and can accommodate
mezzanine Flex-10 or FlexFabric adapters). Each Flex-10 or FlexFabric adapter has four FlexNICs on each of its dual
ports. The multiple FlexNICs can support the iSCSI storage, specialized virtual machine function, management, and
production networks recommended for HP VDI.
VC Flex-10 modules and adapters aggregate Ethernet and accelerated iSCSI storage traffic between the server and Flex-
10 module (server-network edge) into a 10Gb link. Flex-10 technology partitions the 10Gb data stream into multiple (up
to four) adjustable bandwidths, preserving routing information for all data classes.
Flex-10 modules and stacking
You can link the Flex-10 modules within each enclosure, and between enclosures, using the internal links available in
each enclosure, and using stacking cables and appropriately configured ports on each module. This multi-enclosure
stacking creates a single Virtual Connect domain between all the enclosures (up to four enclosures). This means that all
HP BladeSystem VDI servers and the P4800 SAN operate in the same VC domain.
The ability to consolidate all VDI server/storage-side traffic on a single internal VC network, is a major advantage. Any
network adapter can connect with any other adapter within the VC network (domain) without exiting to the external
network, and without additional switches or network management. This simplifies management, reduces support
requirements and a tremendous reduction in both uplink and switch ports. Figure 6 shows a Virtual Connect Flex-10
module.
Figure 6: The HP Virtual Connect Flex-10 Ethernet module




10
Benefits for VDI:
Network isolation Flex-10 technology provides physical network isolation for required network types. VM traffic
within the server has access to eight FlexNICs per dual port Flex-10 or FlexFabric adapter (four FlexNICs per port).
Network aggregation The Flex-10 or FlexFabric adapter aggregates each set of four FlexNICs into a single 10Gb
uplink so that each dual port adapter has only two 10Gb links between the server adapter and the Flex-10 module.
From the enclosure or Virtual Connect domain to the core, as few as two cables aggregate networks you choose to
egress straight to the core resulting in a dramatic reduction in cables as well as expensive switch ports.
Single internal network You can stack Flex-10 modules within the enclosure and between enclosures to create a
single internal VC network (domain), eliminating the need for additional switches and greater intervention by network
administrators, and in the case of the P4800 G2 SAN Solutions for BladeSystem, storage administrators.
Unified management The Onboard Administrator, Virtual Connect Enterprise Manager and LeftHand Centralized
Management Console reduce and simplify the management burden for VDI administrators. Management and
monitoring of the core pieces can be surfaced into your virtualization management software to further simplify the
management picture.
HP 12500 Series Switches
When implemented with HP Virtual Connect, the 12500 series switch facilitates direct wiring to the core saving switch
ports, cables and complexity while flattening the network within the data center. The resultant loss in switch ports,
cables and management translates directly to lower costs, greater security through the reduction in access points and
tremendous performance. Necessary services such as Active Directory, email and application delivery are immediately
accessible.
Figure 7: HP 12500 Series Switch

Extend VDI to the branch with HP MSR series routers
HP Networking has published a reference architecture to help customers understand the implementation and design
requirements as well as the benefits of making the shift to industry standard based networking. Consult the Data Center
Network Reference Architecture at http://h20195.www2.hp.com/V2/GetDocument.aspx?docname=4AA3-4150ENW.



11
HP Thin Clients
Figure 8: HP Flexible Thin Clients

Innovative, reliable, secure. HP delivers real business results for your virtual computing world. We make it easy to
deploy and manage thin client solutions whether youre near or far. HP provides best-fit solutions plus worldwide
support, offering the widest range of choices that allow you to scale up or down. We offer maximum flexibility with
industry-standard hardware and open software platforms, and our technology integration is seamless.
Flexible Thin Clients
Fast, flexible, secure. Give your business what it deserves. Our engineers did the hard work to develop a smart design
with feature-rich power, so you can focus on what matters for you. We know key to growth is staying flexible. We are by
your side to help make those growing pains less painful. We have designed our Flexible Thin Clients with the security you
need now (and the security youll need in the future), with options to grow and expand, all with performance as if you
were on a PC.
Smart Zero Clients
Keeping it quick, simple, and flexible is the goal of HP Smart Zero Clients. No need to worry if the backend software
changes. HPs Smart Zero Clients operate like a zero client, but are intelligent enough to reconfigure themselves to
Citrix, Microsoft, or VMware protocols. Simply plug them in, turn them on, and they work.
Citrix XenDesktop
Citrix XenDesktop is a desktop virtualization solution that transforms Windows desktops and apps into an on-demand
service available to any user, any device, anywhere. XenDesktop quickly and securely delivers any type of virtual
desktops, Windows, web and SaaS (software as a service) applications to PCs, Macs, tablets, smartphones, laptops and
thin clients all with a high-definition user experience. Citrix FlexCast delivery technology enables IT to go beyond VDI
and deliver virtual desktops to any type of user, including task workers, mobile workers, power users and contractors.
XenDesktop also helps IT rapidly adapt to business initiatives, such as offshoring, mergers and acquisitions, and branch
expansion by simplifying desktop delivery and enabling user self-service. The open, scalable and proven architecture
dramatically simplifies virtual desktop management, support and systems integration, optimizes performance,
improves security, and lowers costs.
The best choice for desktop virtualization
XenDesktop is a comprehensive desktop virtualization solution that includes all the capabilities required to deliver
desktops, apps and data securely to every user in an enterprise. Trusted by the world's largest organizations,
XenDesktop has won numerous awards for its leading-edge technology and strategic approach to
desktop virtualization.



12
XenDesktop helps businesses:
Enable virtual workstyles to increase workforce productivity from anywhere
Leverage the latest mobile devices to drive innovation throughout the business
Rapidly adapt to change with fast, flexible desktop and app delivery for offshoring, mergers and acquisitions, branch
expansion and other initiatives
Transform desktop computing with centralized delivery, management and security
A complete line of XenDesktop editions lets you choose the ideal solution for your business needs and IT strategy.
XenDesktop VDI edition, a scalable solution for delivering virtual desktops in a VDI scenario, includes Citrix HDX
technology, provisioning services, and profile management. XenDesktop Enterprise edition is an enterprise-class
desktop virtualization solution with FlexCast delivery technology that delivers the right type of virtual desktop with on-
demand applications to any user, anywhere. The comprehensive Platinum edition includes advanced management,
monitoring and security capabilities.
Provisioning in XenDesktop
Citrix XenDesktop offers several methods of provisioning and deploying desktops. Citrix Provisioning Services (PVS)
infrastructure is a software-streaming technology allowing computers and virtual machines to be provisioned and
managed using a single shared-disk image. All management can be done on a single master image. A snapshot of a
devices hard drive (physical device or virtual machine) is converted into a vDisk (virtual disk) and stored on the PVS
server. This vDisk can then be streamed to PXE booted devices, either physical or virtual.
PVS has write cache files associated with each VM, these files are a defined size and will not grow beyond that set size.
As a best practice, the write cache files should be created as fixed. Also, the files do not have to reside in the same
location as the master image, and for best performance it is recommended to never place the write cache files on the
PVS server (i.e. do not configure for Server-side write cache mode). The write cache files should be maintained either
on a SAN accessible by the hypervisor host or on direct attached storage on the host.
A second method of provisioning VMs is using MCS, or machine creation services. With MCS using the Desktop Studio in
XenDesktop a catalog type of Pooled is created, and an existing VDI VM is selected as the master image. Then the
number of VMs to be created is specified and differential and identity disks for each VM are created on the specified
server or on shared storage for access by multiple servers. The differential files are dynamically created, but can grow to
be the same size as the master image. The identity disks require very little storage and do not change size. It should be
noted that with MCS the differential/identity disks and the master image must reside on the same storage repository. If
there are 10 servers and the master image is placed on each server, there are now 10 master images to manage. If the
master image is placed in a cluster and shared by the 10 servers there is only one image to manage, but all of the
differential/identity disks are in the same storage location. If each server is supporting 100 users, there are now 1000
differential files being read/written to the shared storage, and any or all of these differential files could grow to the size
of the master image.
PVS is more involved to setup and configure, requiring either a physical or virtual server to install PVS, and depending on
the size of the environment may require multiple PVS servers. However, due to the performance and flexibility that PVS
offers, this RA will focus on PVS to do provisioning for the VDI sessions.
Something new in XenDesktop 5.6 is the personal vDisk, or PvD. This feature of XenDesktop retains the option of using a
single image while allowing users to customize their desktop environment and maintain those customizations across
logins, creating a persistent user but managing a single image. PvD uses redirection of changes made in a users VM to a
separate vDisk attached to the users VM. The PvD has two parts using different drive letters and are by default equal in
size:
The first part comprises the C:\Users folder in Windows 7 containing user data, documents, and the user profile.
The second part is a Virtual Hard Disk (VHD) that contains other items, for example programs installed into
C:\Program Files. This drive is hidden from the users and not displayed in Windows Explorer.
How to configure and manage PvD can be found at citrix.com/edocs.
The addition of a PvD to a VM does not increase the number of IOPs associated with a user, but redirects the IOPs
between the streamed image, the write cache file, and the PvD in a PVS configuration. The current version of PvD does
have a CPU impact of roughly 20% on the host. This needs to be considered when looking at how many users can be
supported with PvD.
More information about XenDesktop, PVS, MCS and PvD can be found citrix.com/edocs.



13
Microsoft Windows Server 2008 R2
Microsoft Windows Server 2008 R2 SP1 was used to supply the Hyper-V hypervisor. This hypervisor was used for host
VMs on all of the blades. The Dynamic Memory capabilities of Windows Server 2008 R2 SP1 was utilized to balance and
distribute memory between the VMs. All VMs were started with a minimum of 768 MB of memory and a maximum of
1.5GB.
Microsoft Windows Server 2008 R2 offers the capability to do live migration of VMs using Clustered Shared Volumes
(CSV). The ability to live migrate is configured for the management servers and the infrastructure VMs running on those
servers. Live migration was not implemented for the desktop VMs. The desktop VMs are configured with differential files
to hold changes and updates. These files are deleted and re-created at every login, and are kept on storage directly
attached to the servers. Storing throw-away differential files on expensive SAN storage just so a user can be live
migrated from one host to another needs to be considered in the design. This RA uses direct attached storage to store
the differential files associated with PVS. However, a SAN can be used, but at a higher per user cost for storage and more
complex management infrastructure. It becomes a math problem to ensure the SAN has enough IOPs to support the
users. In the event of a system failure, the user simply logs back in and is connected to a VM running on a different host.
Using Citrix maintenance mode and planning will allow for users to be moved from one server to another by simply
having them log off and back on. If the environment is configured correctly the user will see no changes, nor be aware
they are on a different server and VM. For the infrastructure VMs it is recommended to configure for live migration. This
will be discussed in more detail later in the document.
System Center 2012 Virtual Machine Manager
System Center 2012 is offered as a suite. The Virtual Machine Manager component of System Center 2012 manages the
virtual machines and interfaces with the Citrix Desktop Delivery Control to start, stop, delete, and create virtual
machines. A new set of features are included with System Center 2012 Virtual Machine Manager:
Storage Management
Bare-metal Provisioning
Hyper-V Clustering
Server update services
It is also important to note that with these new features also comes new increased VM scalability with VMM 2012 and
XenDesktop. Microsoft and Citrix jointly tested VMM 2012 and XenDesktop 5.6 at the Microsoft Enterprise Engineering
Center in Redmond, Washington. This project resulted in a new certified scalability architecture that shows a single VMM
2012 instance can support 2000 XenDesktop virtual machines. This increase in scalability allows for simpler VDI
management and less VMM instances to maintain.
For more information around Microsoft Windows Server 2008 R2, System Center, System Center 2012 Virtual Machine
Manager, and Microsoft cluster, go to microsoft.com.



14
Reference Architecture overview
Core design
The HP Reference Architecture tries to take into account the software, hardware and communication mechanisms
required to connect the end user with the compute resource as well as with the world at large. As such, HP is uniquely
positioned in the market as a provider of all of the products required to make VDI work from end to end.
Figure 9: VDI Overview

Understanding the VDI Client Virtualization Reference Architecture
VDI is one possible implementation of the Client Virtualization (CV) Reference Architecture. Citrixs FlexCast model offers
several methods of delivering a desktop to a user including hosted desktops. With VDI, a desktop is created as a virtual
machine. Applications and user personality are injected into the core desktop virtual machine and a brokering
mechanism manages connecting end users to the VM via a connection protocol.
More than simply a virtual machine, the runtime VDI instance is the real time compilation of the end users data,
personal settings and application settings with a core operating system instance and shared generic profile where the
applications are either installed locally as a fully installed or packaged instance or streamed from outside the VM. While
seemingly complex at first glance, the proper management of these individual pieces yields a far more flexible end user
computing environment that is simpler to manage. This is accomplished in part by the decoupling of the user and
application from the core operating system. The single OS image with few applications is easy to manage as part of an
overall, comprehensive client management strategy using tools such as HP Client Automation. Once the user and
application are segmented from the device, the device itself starts to matter less as well. With the right approach, an
end user can receive the same experience regardless of what core compute resource they are logged onto at the
moment. This also means that changing the OS layer is greatly simplified making tasks like Windows operating system
migrations simpler for users who have been virtualized. Figure 10 highlights the compute device at runtime in more
detail.



15
Figure 10: The VDI runtime instance

Citrix User Profile Manager can be used to manage the user and application settings, Citrix XenApp can be used to deliver
applications to the user. Figure 11 highlights the overall architecture at a high level.
Figure 11: VDI within the overall enterprise

Figure 12 shows the architecture from the network perspective. There are four internal networks and one external
network that must be considered when utilizing the HP architecture. The external network is the Internet itself which
enables off campus users to access a compute resource. The first internal network is the production WAN/LAN to enable
corporate users to achieve an end user experience. A management network is also in place to support the management
of the hardware via a variety of mechanisms including HP Integrated Lights-Out (iLO). A network is created for the
PXE/PVS boot and streaming traffic. Finally, a storage network is defined to allow communication between initiators and



16
targets. The storage network is only required when doing persistent users and SAN storage. One of the key
differentiators for this architecture when compared against other offerings is the emphasis on network flexibility and
cost for the core of the environment. This is accomplished by migrating as much of the infrastructure and compute
resource as possible to within the Virtual Connect domain. This eliminates or minimizes the need for switch ports, allows
for the management of intra-domain traffic by a single infrastructure administrator and improves performance and
reliability by utilizing mostly cable free internal connections between hosts and management services. Both the
PXE/PVS and Storage networks are completely internal to the Virtual Connect domain.
Figure 12: The HP client virtualization Reference Architecture for VDI with an emphasis on networking

If we look at the previous figure 12, it is clear that Virtual Connect plays a critical role in the overall architecture. Figure
13 is a closer view of how traffic flows and is segmented between the server and the core. Of particular note is the
minimized number of uplinks. A Virtual Connect domain is cabled directly to the core with a minimal number of cables
while maintaining required bandwidth and availability. This eliminates the need for rack switches which not only saves
cost, but also greatly simplifies the management of the network. In the figure below, production traffic and
management traffic that need to reach the network core are sent via shared uplinks. Of equal interest are the internal
networks created. These networks, labeled storage and PXE/PVS in the figure below, traverse between enclosures but
never egress the Virtual Connect domain and thus never touch core network infrastructure. In a VDI environment where
numerous teams may need to interact simply to keep the environment up and running it can be difficult to manage the
various levels of interactions. The internal storage and PXE/PVS networks can reduce or even eliminate involvement
from the storage and network teams. This in turn centralizes ownership of the solution stack and increases flexibility.
One final note is that every ProLiant BL460c Gen8 server blade is equipped with at least two (2) onboard 10Gb FlexNICs
each allowing for the creation of up to four (4) redundant networks of variable bandwidth between 100Mb and 10Gb.



17
Figure 13: Networks internal and external to the Virtual Connect domain




18
In the context of VDI, there is an additional layer that must be comprehended. Within VDI environments, each hypervisor
host communicates through an abstracted virtual networking layer that allows the individual VDI and management VMs
to communicate with the production network as well as the individual hypervisor hosts to transmit traffic to storage, be
managed and make use of any special function networks. Figure 14 highlights the relationship between the server,
hypervisor, virtual machines, virtual networking layer, FlexNICs and Virtual Connect.
Figure 14: Virtual networking stack from FlexNIC through to the virtual machine

Persistent and non-persistent VMs
A VDI user is considered to be either a persistent or a non-persistent user. A persistent user maintains changes within
the VM across logins. To support persistent VMs normally dedicated VHD files are assigned to each user or a set of
differential files are assigned to a user. Any changes the user makes are stored in these files, and remain there for the
next time the user logs in. With dedicated VHD files, this can be extremely storage intensive. If each user has a 50GB VHD
file assigned to them, and there are 1000 users, then 50000 GB, almost 49 TB, of disk space is required. If using
differential files, these files can grow to be as large as the base image file over time and the potential of users losing
everything in the differential files exists if care is not taken in maintaining and updating the master image file and the
differential files.
For a non-persistent user a single image is shared by all users, and this image is write protected. Any changes or
modifications the user makes while logged in are stored in differential files. These changes made while logged in will be
lost on log out, and at the next login the user will be presented with the original pristine VM and desktop. This does not
allow for user customization of their environment. For non-persistent users, user virtualization needs to be



19
implemented using tools like AppSense
4
or Citrix User Profile Manager. This allows a user to customize their
environment and have those customizations applied at each login.
With the addition of Personal vDisk in XenDesktop 5.6, Citrix has provided a method to use non-persistent single image
management but allow the user the benefits of being a persistent user.
For this paper, all users will be treated as non-persistent users sharing a single or multiple images with Citrix User
Profile Manager or AppSense used to allow each user to customize their environment. However, not all users fit into a
non-persistent model, there are users that require persistence due to application requirements, storage requirements,
or other factors. For these users, a Personal vDisk will be assigned, giving this set of non-persistent users a persistent
environment. The PvD will be stored on a SAN to allow the user to login to any server in the environment.
The Reference Architecture
Whether hosting a 100 user proof-of-concept or a 20,000 user production implementation, the building block of the HP
Reference Architecture for Client Virtualization remains the same. To understand the different models and how they fit
together, you must understand the testing and characterization per server.
Testing methodology
HP has a longstanding tradition of using its own tools to measure server performance for VDI. In the past, these
methods have produced reliable numbers that were achievable in real world usage. This greatly helped with planning
VDI implementations and setting realistic expectations around costs. There has been a demand for a standardized
number that reflects the capacity of a system in addition to a reliable estimation of real world performance.
The industry as a whole has settled on more of a benchmark approach using a small number of tools to approximate
system capacity. HP shunned this approach for a period of time as results produced tended to be extremely optimistic
and when misused resulted in much higher costs for VDI at implementation time versus estimation time. HP sought a
tool that could thus approximate both capacity and real world numbers. After examining tools, HP settled on Login VSI
from Login Consultants.
For testing, HP used the Login VSI 3.5 test tool. Login VSI measures response times for opening applications and clicking
within the UI of applications. The application stack utilized includes Microsoft Word, PowerPoint, Excel, Internet Explorer
and Adobe Flash video. Testing was done running the light, medium, and heavy loads as defined by Login VSI. The
results of the Login VSI testing are given as VSIMax, the point at which response times exceed the acceptable limit. With
Login VSI, VSIMax can be either a set value of 4000ms referred to as VSIMax Classic, or a dynamically calculated value
using a baseline response within the test referred to as VSIMax Dynamic. All VSIMax numbers in this document are
VSIMax Dynamic at the recommendation of Login VSI. Information about Login VSI and details about administering the
test environment can be found at loginvsi.com.
Test environment
All tests were performed using the XenDesktop configuration defined in this document. All infrastructure components
were virtualized and running on a single management server, no redundancy was created for testing purposes. The
server under test was an HP ProLiant BL460c Gen8 with 2 Intel Xeon E5-2680 processors, Hyper-Threading was on, and
256 GB of 1333MHz PC3-12800R RAM. HP testing with Hyper-V has shown that memory speed has a major impact on
performance, with 1333 speeds showing over a 15% increase in number of users over 800 speeds, when using the same
amount of memory.
The virtual machines were optimized using the PVS image conversion tool and additional optimizations recommended
by Citrix in their Windows 7 optimization guide
5
. The virtual machines were configured with Windows 7 x64 SP1 and a
single vCPU. Microsoft Dynamic Memory was utilized, with each VM having a minimum of 768 MB of memory and a
maximum of 1.5GB. This configuration was used for all tests.

4
More information about AppSense can be found at appsense.com
5
Optimization guide can be found at: http://support.citrix.com/servlet/KbServlet/download/25161-102-648285/XD%20-
%20Windows%207%20Optimization%20Guide.pdf



20
Workloads
Login VSI offers a number of workloads intended to emulate different user behaviors. To best emulate HPs traditional
test methodology, the medium, light, and heavy user workloads were chosen. These workloads represent productivity,
task, and knowledge workers respectively.
Detailed information about Login VSI is available at loginvsi.com. An explanation of how tests are run as well as how
scores are calculated is available at loginvsi.com/en/admin-guide.
For the tests HP started the launchers at a fixed, 30 second interval. There were three physical launchers and 25 virtual
launchers. All launchers are connected to a domain and configured for NTP services. Tests are run multiple times to
validate consistency as well as to check for problems in scripts. The best result of all tests is reported in this document.
Because a response time value is used to determine success it is extremely difficult to re-create the exact test results
each time. There are many influences that could impact the results. No test result varied by more than 3-5% between
different runs of the exact same test. Prior to the start of any test run the server under test (SUT) is rebooted and all
virtual machines are restarted. The SUT is allowed a period of rest to reclaim memory prior to test start.
Results
For the medium user workload, VSIMax showed a value of 214 as shown in figure 15.
Figure 15: Login VSIMax and response times for Medium user test.




21
Prior to the tests starting the overall idle time for the SUT for all logical processors was over 80%, with total logical
processor idle time averaging 90-95%. When VSIMax was reached, idle time for the logical processors went to nearly
zero, with LP0 (logical processor 0) being flat lined at zero idle time. LP0 having zero idle time means the system has
been saturated. The following graph highlights the Total idle time for all logical processors (higher black line) and LP0
(lower black line).
Figure 16. Logical Processor Idle Time.

Changing the workload to Light workload a VSIMax was not achieved, in other words response times did not exceed an
acceptable limit. The final tests had 265 VMs running in 256 GB of memory on the server. All 265 VMs were active and
running under load, memory allocation on the server was near 100% and Microsofts dynamic memory was beginning to
come into play. The logical processors had not reached zero idle time, holding just under 10% idle.



22
For the heavy user workload VSIMax was 174 as shown in Figure 17.
Figure 17: Login VSIMax and response times for heavy user test

During testing flash offload using Citrix HDX 3 was utilized. This allows any flash video to be executed on the endpoint
device, if that device has enough compute power to handle running the video. Testing in the lab showed offloading of
flash video gained almost 10% in performance for the server over server side rendering. As noted, the end point device
must be powerful enough to support rendering the flash video, otherwise the rendering will occur on the server.
In all cases LP0 idle time was at almost a constant zero when VSIMax was achieved.



23
Real world vs. lab
All numbers given here were achieved in the lab under known circumstances, including exactly the workload being
executed. The performance you achieve may/will be different depending on the workloads of the users. HP recommends
doing an assessment of your environment to determine the workload, types of users, and which users are best suited for
going to VDI. HP offers an assessment program called CVAM (Client Virtualization Analysis and Modeling). For more
details on CVAM go to http://h20195.www2.hp.com/V2/GetDocument.aspx?docname=4AA3-2409ENW or contact your
HP representative for more details. Adherence to best practices during all aspects of the design and build phases of your
VDI implementation as well as choosing optimized methods for deploying and managing VDI will nearly always result in
greater user density. Selecting problematic applications, poorly suited user groups and failing to follow
recommendations will nearly always have the opposite effect. This is critical because these densities have a potentially
large impact on the acquisition and long term costs of running VDI in a given environment.
From HPs viewpoint, Login VSI is an excellent capacity analysis tool. When you achieve Login VSIMax the user
experience is no longer acceptable. However, the user experience is affected prior to the maximum point. If you look at
the VSI Index Average in the VSIMax graphs there is an inflection point where the line starts to trend upward, estimated
at 60-70%. At this point HP feels the user experience will begin to deteriorate, and this is a more realistic number of
users for the defined workloads. For light and medium level users a range of 60-65% is used, for the heavy level user the
range is 55-60%. Table 1 shows the recommended numbers.
Table 1. Recommended planning numbers by user type.
User Type VSIMax
Recommended Range
Client side flash rendering
Recommended Range
Server side flash rendering
Medium 214 128-139 116-126
Heavy 174 95-104 86-94

As noted previously, HP did not achieve a VSIMax with light user workload running 265 VMs and client side flash
rendering. When doing server side flash rendering a VSIMax of 236 was achieved.
There are many variables that will affect your sizing, from application delivery, operating system, protocols, and
provisioning to number of active users and total number of users. As stated before HP recommends doing an
assessment of your environment as part of the planning process for VDI.



24
The non-persistent model
Figure 18 highlights a single BladeSystem enclosure and four D2700 disk enclosures supporting non-persistent users,
with a user count of 1690 using 130 users per blade. As stated previously, this document will not provide step by step
configuration details.
Figure 18: Non-persistent Configurations.




25
Table 2 represents a bay by bay placement of each component in the configuration. A device bay (D) refers to the bays
on the front of the c7000 enclosure that hold servers and storage. These bays are numbered left to right and top to
bottom with the top row being labeled bays 1 through 8 and the bottom row labeled bays 9 through 16. An interconnect
bay (I) refers to the numbered bays in the rear of the c7000 enclosure that handle I/O communication for the
environment. The Onboard Administrator (OA) bays are assumed to be filled with redundant OAs and are not listed in the
table.
HP D2700 7.2 TB disk enclosures provide direct attached storage to each of the servers. Six drives in a RAID 1+0
configuration are assigned to each server using the SAS Manager within the OA of the BladeSystem enclosure. The disk
enclosures are daisy chained together. A VSA is on each management server to leverage the D2700 drives assigned to
each server, creating a cluster between the management servers.
Table 2. Device by bay and function
Bay Device Function
D1 HP ProLiant BL460c Gen 8 Hosts core management VMs Broker, SQL, DHCP for PXE Network, PVS, VSA
D2 HP ProLiant BL460c Gen 8 Hosts end user computing VMs
D3 HP ProLiant BL460c Gen 8 Hosts end user computing VMs
D4 HP ProLiant BL460c Gen 8 Hosts end user computing VMs
D5 HP ProLiant BL460c Gen 8 Hosts end user computing VMs
D6 HP ProLiant BL460c Gen 8 Hosts end user computing VMs
D7 HP ProLiant BL460c Gen 8 Hosts end user computing VMs
D8 HP ProLiant BL460c Gen 8 Hosts end user computing VMs
D9 HP ProLiant BL460c Gen 8 Hosts core management VMs Broker, SQL, DHCP for PXE Network, PVS, VSA
D10 HP ProLiant BL460c Gen 8 Hosts end user computing VMs
D11 HP ProLiant BL460c Gen 8 Hosts end user computing VMs
D12 HP ProLiant BL460c Gen 8 Hosts end user computing VMs
D13 HP ProLiant BL460c Gen 8 Hosts end user computing VMs
D14 HP ProLiant BL460c Gen 8 Hosts end user computing VMs
D15 HP ProLiant BL460c Gen 8 Hosts end user computing VMs
D16 HP ProLiant BL460c Gen 8
Hosts core management VMs Additional PVS VMs to support increased number of
users, VSA
I1 HP Virtual Connect Flex-10 Module Master storage and Ethernet traffic communication module
I2 HP Virtual Connect Flex-10 Module Redundant storage and Ethernet traffic communication module
I3 HP 6Gb/s SAS Switch SAS link for connecting DAS storage to blades
I4 HP 6Gb/s SAS Switch SAS link for connecting DAS storage to blades




26
There are three management servers in the configuration. Each management server runs the infrastructure VMs
required to support the XenDesktop configuration.
Management VM configurations
The infrastructure VMs are defined in the next table. Microsoft Clustering Shared Volumes is used between the
management servers to allow migration of the infrastructure VMs.
Table 3. Management VM configuration information
Virtual Machine Function Operating System
Number
of VMs
vCPU Memory
Disk
Size
PXE DHCP Microsoft Windows Server 2008 R2 SP1 Datacenter 1 2 6GB 40GB
PVS Microsoft Windows Server 2008 R2 SP1 Datacenter 4 4 16GB 50GB
VMM 2012 Microsoft Windows Server 2008 R2 SP1 Datacenter 2 2 8GB 50GB
Citrix Desktop Delivery
Controller(DDC)
Microsoft Windows Server 2008 R2 SP1 Datacenter 2 4 8GB 40GB
SQL Server Microsoft Windows Server 2008 R2 SP1 Datacenter 2 2 8GB 60GB
HP Systems Insight Manager 7/
LeftHand Centralized
Management Console
Microsoft Windows Server 2008 R2 SP1 Datacenter 1 2 8GB 60GB
HP LeftHand VSA HP SAN/iQ 9.5 3 2 4GB NA

Each management server is assigned 6 drives from the D2700. These drives are configured with RAID 5 for increased
storage. These drives are assigned to the HP VSA running on each management server. A SAN cluster is created using the
LeftHand CMC (Central Management Console) between the VSAs. This will create a 1500GB cluster for two management
servers, 2000GB for three management servers. Within this cluster, seven volumes are created. The following table
defines the volumes.
Table 4. Cluster Volumes
Volume Name Access by Size
Master Management servers, holds VM vhd files 500GB
PVS_1 PVS Server 1, holds master image files 250GB
PVS_2 PVS Server 2, holds master image files 250GB
SQL_Volume SQL VMs 250GB
VMM-Library VMM VMs for library 250GB
PVS_3 PVS Server 3 250GB
PVS_4 PVS Server 4 250GB




27
For PVS, 400-500 streams is the HP recommendation for a virtualized PVS server. The breakdown on what VMs run on
which server by default is shown in the next table.
Table 5. Infrastructure VM and Management Host
Server VMs
Management Server 1 PXE DHCP, PVS, DDC, VSA, SQL, SC 2012 VMM
Management Server 2 PVS, DDC, VSA, SQL, SIM/CMC, SC 2012 VMM
Management Server 3 Two PVS VMs, VSA

The VSAs are stored on the internal drives of each management server. The DAS storage associated with each
management server is given to the VSA to manage, and all VSAs are put into a cluster. The management servers attach
to the VSA cluster to store the other infrastructure VM VHD files. This allows clustering to be setup between the
management servers, allowing for live migration of the infrastructure VMs between the management servers. The other
VSA volumes are then attached to the correct infrastructure VMs.
For networking, four networks are defined with Virtual Connect as discussed previously. Table 6 lists the networks and
associated bandwidth.
Table 6. Virtual Connect Networks
Network Type Link Speed
Management External 500 Mb
Production External 2 Gb
PXE/PVS Internal 5.5 Gb
Storage Internal 2 Gb

The Storage network is used for the clustering and migration of infrastructure VMs in the non-persistent model, and in
the persistent model (covered next) to also access the SAN.
Each network is redundant. The BL460c Gen8 FlexNICs have two 10Gb ports that can be divided into 4 NICs each, sharing
10Gb of bandwidth across the 4 NICS.
A rack solution
With a single enclosure supporting 1690 users, two complete configurations can fit into a single rack, creating support
for more than 3380 users. A single Virtual Connect domain can be created between the two enclosures to make the
environment easier to manage, and up to four enclosures can be configured into a Virtual Connect domain using two
racks and supporting over 6500 users.



28
Figure 19: Two full racks Over 6500 users

The persistent model
As discussed before, the persistent model maintains changes to the VM over logins. For all users, persistent or non-
persistent, a user virtualization tool like Citrix Profile Manager should be used to manage the user profiles and
customization. A persistent user has needs beyond that of other users, such as needing to install specific software that
you do not want in the master image or streamed from a XenApp or AppV server. This user has requirements/needs
beyond the average user. This personal vDisk should not be used to store user data, as with non-persistent users the
user data should be stored on a network device like a CIFS share. The personal vDisk will be stored on a SAN added to the
hardware. This prevents the persistent user from being tied to a single physical server. To accomplish this a catalog of
servers is created in XenDesktop Desktop Directory to connect to the P4800 SAN and support the persistent users.
In most customer situations the non-persistent/persistent mix runs about 80/20, but for this RA a 50/50 ratio is used. To
allow the persistent user to not be locked to a specific server, the PvD will be stored on an HP LeftHand P4800.
To accommodate the P4460sb blades of the P4800, two Hyper-V hosts must be removed, so in a fully configured
enclosure for persistent users, the limit is 1170 users. With less VDI hosts, the number of required D2700 storage
enclosures is reduced to 3. Because PvD is being used to support the persistent users, and as stated earlier, CPU
utilization is impacted and a performance reduction is incurred. This performance impact is calculated at 20%. Since half



29
of the users are considered to be using PvD, then the number of users supported on blades with PvD users is reduced to
104 users (130 20%). Total number of supported users is 5 servers * 130 user + 5 servers * 104 users equaling 1170
users.
Figure 20: Persistent Configuration, 50/50 user mix.

The changes are bays 7 and 8 now have the P4460sb blades in them and bay 15 is empty, the P4800 storage and
switches have been added, and only three D2700 storage enclosures are required. The following table lists the
device/bay configuration for the rack.
Table 7. Device by bay and function
Bay Device Function
D1 HP ProLiant BL460c Gen 8 Hosts core management VMs Broker, SQL, DHCP for PXE Network, PVS, VSA
D2 HP ProLiant BL460c Gen 8 Hosts end user computing VMs



30
Bay Device Function
D3 HP ProLiant BL460c Gen 8 Hosts end user computing VMs
D4 HP ProLiant BL460c Gen 8 Hosts end user computing VMs
D5 HP ProLiant BL460c Gen 8 Hosts end user computing VMs
D6 HP ProLiant BL460c Gen 8 Hosts end user computing VMs
D7 HP P4460sb Storage Blade Storage controller one for the HP LeftHand P4800 SAN for BladeSystem
D8 HP P4460sb Storage Blade Storage controller one for the HP LeftHand P4800 SAN for BladeSystem
D9 HP ProLiant BL460c Gen 8 Hosts core management VMs Broker, SQL, DHCP for PXE Network, PVS, VSA
D10 HP ProLiant BL460c Gen 8 Hosts end user computing VMs
D11 HP ProLiant BL460c Gen 8 Hosts end user computing VMs
D12 HP ProLiant BL460c Gen 8 Hosts end user computing VMs
D13 HP ProLiant BL460c Gen 8 Hosts end user computing VMs
D14 HP ProLiant BL460c Gen 8 Hosts end user computing VMs
D15 Blank Blank
D16 HP ProLiant BL460c Gen 8 Hosts core management VMs Broker, SQL, DHCP for PXE Network, PVS
NOTE: VSA not required, P4800 will supply storage for third management server
I1 HP Virtual Connect Flex-10 Module Master storage and Ethernet traffic communication module
I2 HP Virtual Connect Flex-10 Module Redundant storage and Ethernet traffic communication module
I5 HP 3Gb/s SAS Switch SAS link for the HP LeftHand P4800 SAN for BladeSystem
I6 HP 3Gb/s SAS Switch SAS link for the HP LeftHand P4800 SAN for BladeSystem
I7 HP 6Gb/s SAS Switch SAS link for connecting DAS storage to blades
I8 HP 6Gb/s SAS Switch SAS link for connecting DAS storage to blades

NOTE: Additional 3Gb/s SAS switches are added to the back interconnects to support the P4800. The P4800 requires
separate 3Gb/s SAS switches, which by default are placed in interconnect bays 5 and 6. The P721m for the Gen8 servers
must reside on the blade mezzanine slot 2, meaning the 6Gb SAS switches needed to connect to the D2700s by default
would also be placed in interconnect bays 5 and 6. The most straight forward solution is to place the 6 Gb SAS switches
in bays 7 and 8, and use the Onboard Administrator to disable access to those bays for the P4460sb blades associated
with the P4800. This also assumes all correct firmware versions for the Onboard Administrator, SAS switches and
blades. Please check with your local service representative for complete HP recommendations.
NOTE: Availability of the P721m card is scheduled for August 2012.
The third management server does not have direct-attached storage and no VSA is necessary. A 500 GB volume is carved
out from the P4800 and added to the existing management cluster. The PVS VMs associated with the third management
server are placed in this volume as part of the cluster using CSV, Cluster Shared Volumes, to allow migration of the PVS
VMs with the other management servers.



31
The P4800 is carved into multiple volumes to hold the PvD files. For this RA, a size of 10GB is used for the PvD, and 250
PvD users are supported for each volume.
A complete rack
A complete rack can be configured to support 2340 users, two racks can be linked together to create a 4 enclosure
Virtual Connect domain which can support over 4500 users. This does assume the 50/50 ratio for persistent and non-
persistent users.
Figure 21: Multiple Persistent Model Racks

HP, Citrix, and Microsoft better together
HP is an alliance partner with both Citrix and Microsoft, and the engineering and marketing teams for all three
companies work together to certify and integrate joint solutions and produce complete solutions for our customers. HP,
Citrix and Microsoft enable you to:
Standardize on a common virtualization platform fully certified at the operating system level by Microsoft
Simplify server, storage, and network management with HP Insight Control for Microsoft System Center
Optimal density of valuable data center floor space using HP ProLiant Gen8 servers



32
Summary
This reference architecture outlines the performance of Citrix XenDesktop 5.6 on Windows Server 2008 R2 using the
Hyper-V hypervisor running on HP ProLiant BL460c Gen8 servers. It also discusses the hardware and software
configuration around the test environment that can be implemented in real world configurations.
Appendix A
The hardware required for configuring the RA is broken into several components. For ease of hardware configuration HP
offers Factory Express to rack and stack the hardware. Using Factory Express will put all of the hardware into the rack
using HP best practices. Contact your HP sales representative for using HP Factory Express in racking your hardware.
For rack and power infrastructure the following table defines the individual components.
Table A1. Rack and Power BOM
Qty Part Number Description
1 BW908A HP 642 1200mm Shock Intelligent Rack
1 BW909A HP 42U 1200mm Side Panel Kit
1 BW930A HP Air Flow Optimization Kit
1 BW932A HP 600mm Jb Rack Stabilizer Kit
2 AF916A HP 3PH 48A NA/JP Pwr Monitoring PDU
2 AF502B HP 2 STK 4X Fixed C-13 Offset WW Kit
4 142257-006 HP 10A IEC 320 C14-C13 4.5ft US PDU Cable
As Needed AF070A HP 10pk Carbt 1U Universal Filler Panel

A single rack can support two enclosures and configurations as show in previous figures.
The enclosure configuration is in the next table.
Table A2. Enclosure BOM
Qty Part Number Description
1 TC277AAE HP Insight Control Encl Bundle 16 E-LTU
1 456204-B21 HP BLc7000 DDR2 Encl Mgmt Option
1 507019-B21 HP BLc7000 CTO 3 IN LCD ROHS Encl
1 517520-B21 HP BLc 6X Active Cool 200 FIO Fan Opt
1 517521-B22 HP 6X 2400W Plat Ht Plg FIO Pwr Sply Kit
1 413379-B21 HP BLc7000 1 PH FIO Power Module Opt



33
Qty Part Number Description
1 433718-B21 HP BLc7000 10K Rack Ship Brkt Opt Kit
1 591973-B21 HP VC Flex-10 Ent Edn for BLc7000 Opt
2 453154-B21 HP BLc VC1Gb RJ-45 SFP Opt Kit NOTE: This is optional.
2 455883-B21 HP BLc 10Gb SR SFP+ Opt NOTE: You may choose a direct cable connection for 10Gb connections. Consult
your sales representative.
4 C7535A HP Ethernet 7ft CAT5e RJ45 M/M Cable

For the servers, the following BOM is per server. For the non-persistent configuration 16 servers per enclosure, for the
persistent model 13 servers are required.
Table A3. Single Server BOM
Qty Part Number Description
1 641016-B21 HP BL460c Gen8 10Gb Flb CTO Blade
2 652564-B21 HP 300GB 6G SAS 10K 2.5in SC ENT HDD
1 662063-B21 HP BL460c Gen8 E5 2680 Kit
1 662063-L21 HP BL460c Gen8 E5 2680 FIO Kit
16 672631-B21 HP 16GB 2Rx4 PC3-12800R-11 Kit
1 650072-B21 HP Smart Array P721M/2G FBWC Ctrlr
(Available after August 2012)
1 684211-B21 HP Flex-10 10Gb 2P 530FLB FIO Adptr
1 TA688AAE HP P4000 VSA One per Management server, three management servers for full enclosure
NOTE: if doing persistent with P4800, only two are required

For the D2700 storage, the following part numbers are required:
Table A4. D2700 BOM
Qty Part Number Description
1 BK764A 6Gb SAS Dual Pack one per BladeSystem enclosure
1 AW525A HP D2700 300GB 6G SAS SFF 7.5TB Bundle One for every 4 servers
2 407339-B21 HP Ext Mini SAS 2m Cable need two per D2700




34
SAN storage BOM for P4800:
Table A5. P4800 BOM
Qty Part Number Description
2 AJ864A HP 3Gb SAS BL-c Switch
2 142257-006 HP 10A IEC320 C14-C13 4.5ft US PDU Cable
1 BV932A HP P4800 G2 42TB SAS SAN BladeSystem

For more information
To read more about HP and client virtualization, go to hp.com/go/cv
To read more about HP and Citrix, go to hp.com/go/citrix
To read more about Citrix XenDesktop, go to citrix.com/XenDesktop
To read more about Microsoft Windows Server 2008, go to microsoft.com/windowsserver2008
To read more about Login VSI, go to loginvsi.com
To help us improve our documents, please provide feedback at
http://h71019.www7.hp.com/ActiveAnswers/us/en/solutions/technical_tools_feedback.html.
Call to action
Contact your HP representative today to engage and discover how this client virtualization reference architecture can
unleash your workers productivity, while also streamlining management, better protecting your information and
lowering operating costs.



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 omissions 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. Adobe is a trademark of Adobe Systems Incorporated.
4AA4-2097ENW, Created June 2012