Why HP 3PAR StoreServ Storage for Client Virtualization and best practices

Transcription

Why HP 3PAR StoreServ Storage for Client Virtualization and best practices
Technical white paper
Why HP 3PAR StoreServ Storage
for Client Virtualization and best
practices
Table of contents
Executive summary ...................................................................................................................................................................... 2
Introduction .................................................................................................................................................................................... 2
Key HP 3PAR StoreServ client virtualization advantages ................................................................................................ 3
Client virtualization overview ................................................................................................................................................. 3
Client virtualization and storage considerations .................................................................................................................... 4
HP Client Virtualization Analysis and Modeling Service .................................................................................................... 4
Mixed workload support .......................................................................................................................................................... 5
Performance .............................................................................................................................................................................. 5
Capacity ....................................................................................................................................................................................... 6
Resilience .................................................................................................................................................................................... 6
Sizing and scaling ...................................................................................................................................................................... 7
Persistent and non-persistent implementations .............................................................................................................. 9
Cost .............................................................................................................................................................................................. 9
Setting up and configuring HP 3PAR StoreServ arrays for client virtualization ............................................................ 10
Common provisioning groups .............................................................................................................................................. 13
Thin versus full allocations ....................................................................................................................................................... 17
Full provisioning ...................................................................................................................................................................... 17
Thin provisioning ..................................................................................................................................................................... 17
HP 3PAR StoreServ Flat SAN and Direct Attached Storage ................................................................................................ 18
3PAR file services and Network Attached Storage (NAS) ................................................................................................... 19
Disaster Recovery: Peer Persistence ...................................................................................................................................... 20
Best practices for optimization ................................................................................................................................................ 21
HP 3PAR Dynamic Optimization Software......................................................................................................................... 21
HP 3PAR Adaptive Optimization Software ........................................................................................................................ 22
Client virtualization products................................................................................................................................................ 22
Summary ....................................................................................................................................................................................... 22
Appendix A: Host Persona ALUA Support for ESX ................................................................................................................ 23
Appendix B: Effect of read caching on storage performance ............................................................................................ 24
Appendix C: Configuring HP 3PAR StoreServ Snapshot Administrator (SA) space ....................................................... 25
Acronym list .................................................................................................................................................................................. 25
For more information ................................................................................................................................................................. 26
Technical white paper | Why HP 3PAR StoreServ Storage for Client Virtualization and best practices
Executive summary
Solution architects and IT managers face many challenges in delivering powerful, secure and reliable end user computing
capabilities with today’s data center. Relentless demands on infrastructure, combined with limited resources and escalating
security and maintenance concerns increase pressure on managers and users alike. The HP 3PAR StoreServ storage
solution offers a client virtualization approach that addresses these and many other critical concerns.
Replacing the physical desktop with a virtual client simplifies maintenance by enabling the central management of updates
and patches for hundreds or thousands of clients simultaneously. In addition, using virtual clients on centrally managed
storage within a secure data center protects your data from unwelcome intrusions.
Autonomic management automates, simplifies, and expedites storage management operations such as storage
provisioning, tiering, and change management without administrator intervention.
The HP 3PAR StoreServ storage solution for client virtualization detailed in this paper offers simplified, yet sophisticated
management tools and data protection to address data storage concerns for client virtualization deployments of all sizes
and degrees of complexity. HP 3PAR StoreServ storage offers effortless performance and flexibility needed to accelerate
your client virtualization deployment for today’s needs and tomorrow’s growth, no matter how varied.
HP 3PAR StoreServ storage has the architecture to scale with your growing needs—regardless of whether you are looking
to implement a proof of concept or a full-scale production environment to support tens of thousands of clients.
The information presented in this paper will outline the architecture of the HP 3PAR StoreServ while also describing the
features of the array that are best suited for client virtualization. This paper will not focus on client virtualization
configurations or solution recommendations, but rather on how the HP 3PAR StoreServ benefits client virtualization.
Target audience: IT administrators and solution architects planning or deploying a client virtualization solution with HP
3PAR StoreServ storage.
The reader should have a working knowledge of client virtualization using a Virtual Desktop Infrastructure (VDI).
Introduction
Client virtualization is a solution used by IT organizations to manage escalating demands on infrastructure and diminishing
budgets. The defining feature of client virtualization solutions is the ability to move the user experience and user data closer
together within the secure environment of the data center. Client virtualization alleviates several major concerns for IT
administrators:
• Infrastructure and maintenance complexity
• Mobile and remote access needs
• Security and disaster recovery
• Cost containment directives
• User productivity requirements
In the face of so many competing demands, many IT organizations find the tradition of tying applications and devices to
employee desktops is at the root of their IT shortcomings.
Client virtualization replaces the physical desktop with a virtual one by running a virtual instance of a desktop operating
system for each user. Client virtualization provides the same manageability and availability advantages as server
virtualization, including load balancing and streamlined maintenance. Client virtualization also uses virtual applications,
eliminating the need to apply patches and updates to every operating system on every physical desktop in the enterprise.
Client virtualization enhances data security by moving the user’s data and settings to the data center, which decreases the
chance of theft and unauthorized access. If server performance becomes an issue or a server requires maintenance, client
virtualization allows users to migrate easily to another server.
Client virtualization can pose significant challenges for storage systems, and these challenges can directly determine the
success or failure of any project. Boot storms, login storms, virus scans, and image patching and updating can all drive
tremendous instantaneous I/O requirements on storage systems. These use cases significantly differ in load profile from
typical individual daily client usage requirements for storage. Deploying a storage system that can support the extreme
requirements of boot storms, login storms, virus scans, and image patching/updating while supporting the steady-state
requirements for typical client logins and usage patterns is critical.
HP delivers a comprehensive portfolio of server, storage, and networking Reference Architectures well suited for the diverse
needs of any client virtualization project. This paper focuses on the storage aspects of client virtualization using HP 3PAR
2
Technical white paper | Why HP 3PAR StoreServ Storage for Client Virtualization and best practices
StoreServ storage. HP 3PAR StoreServ storage is an ideal client virtualization storage solution that is designed with
virtualization in mind. HP 3PAR StoreServ storage delivers a platform that provides the performance required for the most
extreme use cases, while delivering the cost efficiency required for everyday use cases.
With a broad range of models, the HP 3PAR StoreServ family, shown in Figure 1, supports a wide variety of client
virtualization projects. HP 3PAR StoreServ products offer value-centric solutions, supplying superior performance and the
flexibility to support client virtualization both now and in the future. HP 3PAR StoreServ also delivers streamlined
management, using advanced autonomic (self-detecting) features without sacrificing critical capabilities such as
performance and scalability for a variety of use cases.
Figure 1: The HP 3PAR StoreServ family
Key HP 3PAR StoreServ client virtualization advantages
Specifically built for virtualization, HP 3PAR StoreServ storage delivers key advantages for client virtualization:
• High performance to meet peak demands, even during:
– Boot storms
– Login storms
– Virus scans
– Image updates and patching
• Non-disruptive scalability to easily support client growth when and where needed
• Bulletproof storage so users are productive at all times, even during outages
• Increased efficiency to ensure no wasted storage
• Effortless storage administration to lower operational costs and reduce time to value
Refer to the HP 3PAR StoreServ Storage product page for more information, hp.com/go/storeserv
Client virtualization overview
Client virtualization is a proven IT strategy that improves employee productivity by supporting mobility and bring-your-owndevice (BYOD) initiatives, while assuring application and data security. This approach improves efficiency and reduces IT
infrastructure management costs. On the surface, the benefits of client virtualization are very attractive. However, without
established deployment methodologies based on storage that is purpose-built for virtualization, these benefits can be
difficult to achieve. HP provides the innovative approaches and solutions needed by IT organizations considering client
virtualization.
Client virtualization currently encompasses four technologies:
• Server-based computing technologies
• Virtual Desktop Infrastructure
• Graphics Enabled VDI
• Blade Workstations
3
Technical white paper | Why HP 3PAR StoreServ Storage for Client Virtualization and best practices
Figure 2 illustrates a client virtualization environment using HP 3PAR StoreServ storage. It shows how a user accesses a
client environment over a display protocol. In this example, the OS combines with the user’s profile, data settings, and
required applications to create a virtualized desktop tied to a user at logon. HP 3PAR StoreServ delivers the back-end
storage needed to support the user’s virtualized desktop.
Figure 2: An approach to client virtualization using HP 3PAR StoreServ Storage
Client Virtualization Infrastructure
User
Display Protocol
User Virtualization
HP 3PAR StoreServ Storage
Application Virtualization
OPERATING SYSTEM
HYPERVISOR
Client virtualization and storage considerations
The following sections describe important considerations for implementing client virtualization using HP 3PAR StoreServ
storage:
• HP Client Virtualization Analysis and Modeling Service
• Mixed workload support
• Performance
• Capacity
• Resilience
• Sizing and scaling
• Persistent and non-persistent implementations
• Cost
HP Client Virtualization Analysis and Modeling Service
Too often, client virtualization is deployed as a technology, without regard to the impact to the users, and ROI for the
business. Completely understanding a client virtualization environment requires intimate knowledge of client usage
patterns. For customers interested in client virtualization, HP offers the HP Client Virtualization Analysis and Modeling
(CVAM) service. Customers are encouraged to work with HP’s client virtualization experts to create an environment that
succeeds at initial deployment.
4
Technical white paper | Why HP 3PAR StoreServ Storage for Client Virtualization and best practices
The HP Client Virtualization Analysis and Modeling Service assists with these tasks:
• Analyzing existing client devices
• Analyzing existing client workload patterns
• Providing storage capacity planning and resource modeling
• Planning for desktop operating system images required to virtualize desktops
• Understanding application usage
• Identifying applications that could pose challenges for a client virtualization project
NOTE: The CVAM process supports all forms of client virtualization, not just VDI. For more information, refer to the HP Client
Virtualization Services web page.
Mixed workload support
It is very rare that any given IT environment supports a single application like client virtualization. In many cases, it is
advantageous to use the same infrastructure to support many different applications, from transactional databases and
decision support systems, to email and collaboration systems, and even Server and Client Virtualization systems. By nature,
client and server virtualization systems aggregate I/O workloads of many different types of applications and produce one of
the most challenging usage profiles. In the industry, this is known as the “I/O Blender” effect, where many different
application workloads are blended into one. This effect can produce a workload characterized by lots of random access
patterns with heavy read and write components.
The HP 3PAR StoreServ architecture features mixed workload support that enables a single HP 3PAR StoreServ array to
support thousands of virtual clients and to house both server and client virtualization deployments simultaneously, without
compromising the user experience. Mixed workload support enables different types of applications (both transaction-based
and throughput-intensive workloads) to run without contention on a single HP 3PAR StoreServ array. This architectural
advantage is particularly valuable in consolidated and virtualized environments, where a single array must reliably support a
wide mix of application types across the cluster while delivering consistently high performance.
Mixed workload support is a part of the architecture of the HP 3PAR StoreServ and requires no user setup.
Performance
HP 3PAR StoreServ storage offers high performance to meet peak demands even during boot storms, login storms, and
virus scans. This architectural advantage is particularly valuable in virtualized environments, where a single array must
reliably support a wide mix of application types while delivering consistently high performance.
Beyond the mixed workload support, discussed earlier, differentiated performance features of HP 3PAR StoreServ storage
that dramatically increase client virtualization environment performance include:
• Wide striping
• Shared cached data
• Pre-fetching
• Write caching
The unique HP 3PAR StoreServ architecture wide stripes volumes across all drives, all controllers, and all ports to deliver
maximum I/O throughput and minimum latencies, which mitigates server memory bottlenecks and traditional storage
constraints. Wide striping provides consistently high levels of performance (regardless of the capacity storage use) during
boot storms, virus scans, and login storms.
Wide striping support is a part of the architecture of the HP 3PAR StoreServ and requires no user setup. Refer to the Logical
components section for an explanation of how wide striping is designed using chunklets.
HP 3PAR StoreServ OS keeps track of read streams for Virtual Storage Volumes (VVs) so that it can improve performance by
“pre-fetching” data from drives ahead of sequential read patterns. Each VV can detect up to five interleaved sequential read
streams and generate pre-fetches for each of them. Simpler pre-fetch algorithms that keep track of only a single read
stream would not recognize the access pattern consisting of multiple interleaved sequential streams.
5
Technical white paper | Why HP 3PAR StoreServ Storage for Client Virtualization and best practices
Pre-fetching improves sequential read performance in two ways:
• Reduces response time seen by the host
• Accesses drives using larger block sizes than the host uses, resulting in more efficient operations. A controller node
caches writes to VVs, mirrors the writes in the cache of another controller, and then acknowledges the host. The host
experiences a shorter response time because writes are acknowledged when the write is mirrored to the second
controller’s cache, instead of waiting for a write to be acknowledged by the disk. This is possible because the mirroring
and power failure handling guarantee the integrity of the cached write data.
In addition to dramatically reducing the host write response time, write caching can often benefit back-end drive
performance by these merging write techniques:
• Merge multiple writes to the same blocks so that many drive writes are eliminated.
• Merge multiple small writes into single, larger drive writes to increase efficiency.
• Merge multiple small writes to a RAID 50 or RAID 6 logical disk (LD) into full-stripe writes, eliminating reads of the old data
for the stripe from the drives.
HP 3PAR StoreServ combined with HP 3PAR OS 3.1.2 or later, is fully compatible with VMware VAAI and Microsoft®
Windows® Server 2012 Offload Data transfer technologies. These technologies offload the processor-intensive tasks (such
as copying hundreds of VMs for desktop provisioning) to the storage array, where they are managed more efficiently;
conserving server and network resources.
Capacity
For a successful client virtualization environment, storage capacity must accommodate virtual machine or differential file
growth (files changed since last master image composition), user data growth, and the addition of new users to the
environment.
The storage infrastructure receives various requests and must load different parts of the operating system and the
computing environment for individual virtual desktops. Properly designing the storage tiering with the appropriate storage
disk drive technology provides the required capacity needed for each virtual desktop.
To ensure maximum utilization of storage capacity, HP 3PAR Thin Persistence software leverages the unique Thin Built In
hardware capabilities of the HP 3PAR StoreServ ASIC to drive the inline “thinning” of storage volumes. This is especially
important to client virtualization environments, where the constant addition and deletion of virtual clients means that even
a thin provisioned volume will eventually become “fat” and inefficient. HP 3PAR Thin Persistence software eliminates this
problem by leveraging built-in, zero-detection capabilities. Administrators are able to maintain optimal storage efficiency by
automatically reclaiming unused space associated with deleted data within a thin provisioned volume.
Resilience
Supporting hundreds to thousands of clients means that the storage deployed for client virtualization must be highly
available. Any outage can cause huge productivity loses for the business. Deploying a highly available storage system is
paramount.
HP 3PAR StoreServ storage features a unique, mesh-active controller clustering architecture, purpose-built for virtual and
cloud data centers. This architecture, shown in Figure 3, combines the benefits of monolithic and modular architectures
while eliminating price premiums and scaling complexities. Unlike legacy “active-active” controller architectures, where each
volume is active on only a single controller, the mesh-active design allows each volume to be active on every node in the
system. This design delivers robust, load-balanced performance and greater headroom for cost-effective scalability,
overcoming the tradeoffs typically associated with modular and monolithic storage. A high-speed, full-mesh
interconnection joins multiple controller nodes to form a cache coherent, active-active cluster.
6
Technical white paper | Why HP 3PAR StoreServ Storage for Client Virtualization and best practices
Figure 3: Illustration of mesh-active controller clustering
HP 3PAR StoreServ Persistent Cache software allows “always on” client virtualization environments to respond gracefully to
unplanned, controller node failures, without the substantial performance penalties associated with traditional arrays and
“write-through” mode. Available in configurations of two or more nodes, this capability is especially important for virtual
client environments, where performance is at a premium during business hours and any reduction in availability can directly
affect hundreds or thousands of users. Persistent cache is also essential to supporting the business-critical applications
deployed as virtual clients.
Mesh-active controller clustering and persistent cache support are parts of the HP 3PAR StoreServ architecture and require
no user setup.
Sizing and scaling
Before implementing your HP 3PAR StoreServ client virtualization design into a production environment, consider these
sizing and scaling issues:
• Number of current and future users in the client virtualization environment
• Types of applications the environment must support
• Number of IOPS each user will produce and storage tiers each I/O will affect
The HP 3PAR StoreServ family provides cost-effective, single-system sizing and scalability through a cache-coherent, multinode, clustered implementation. Modular design provides storage array sizing and scaling to meet client virtualization
demands.
7
Technical white paper | Why HP 3PAR StoreServ Storage for Client Virtualization and best practices
Sizing
Storage sizing requires an understanding of how to convert a traditional environment into a client virtualization
environment.
Sizing requirements depend on several factors. These include:
• IOPS
• Read versus Write ratio
• Type of I/O (Sequential or Random)
Storage bottlenecks and sizing I/O for peak demand
Before sizing the array, consider the use cases. Most IOPS occur during boot and login storms. A best practice is to size
storage based on the peak demand times when users will be logging in or booting up. To ensure the best possible client
experience, different storage technologies may be deployed for different components of a client virtualization environment.
Solid-State storage can be used for centralized golden operating system images to ensure the fastest boot and response
times. Spinning disk may be used for less delay-sensitive client virtualization (CV) data, like user profiles. HP 3PAR StoreServ
storage supports a wide variety of different storage types so you can design the right storage system for your individual
needs.
The solution to the storage I/O bottleneck is to size storage I/O in client virtualization environments for peak demand, not
average I/O utilization. The infrastructure supporting the client virtualization environment should deliver very high IOPS
performance.
A traditional way to boost IOPS is to stripe data across a large number of rotating disk drives. Striping data aggregates the
performance of multiple drives and applies it to the entire data set. However, this is a costly solution because striped drives
use only a small portion of their total capacity, resulting in a large number of poorly utilized drives that consume a
significant amount of power, cooling, and data center space.
HP 3PAR StoreServ provides wide striping built into the architecture of the array. Wide striping provides performance
without the need to short stroke the drives.
Enterprise solid-state drives (SSDs) are a superior alternative to achieve high IOPS. SSDs deliver performance orders of
magnitude higher than disk drives. Enterprise SSDs offer high reliability, low latency, high performance, energy efficiency,
ease of implementation, and long device lifetime.
For high IOPS performance, use the best practice approach of using SSDs as local storage for stateless desktops. This
solution delivers exceptional I/O performance while dramatically reducing storage requirements, increasing server
consolidation, streamlining administration, and lowering client virtualization total cost of ownership.
HP 3PAR StoreServ delivers architectural features to solve these bottleneck issues. Tiered storage, combined with chunklet
provisioning design, provides the performance needed to support all types of client virtualization environments.
Scaling
To accommodate new users and the growth of user data, storage must be scalable. The HP 3PAR StoreServ’s modular
implementation meets the scaling needs for all client virtualization environments.
Figure 4 shows that scaling an existing environment supporting 2,000 users to one supporting 4,000 users only requires
duplicating the existing infrastructure. The modular design of the HP 3PAR StoreServ arrays facilitates scaling and sizing as
the client virtualization environment grows, using existing design plans.
8
Technical white paper | Why HP 3PAR StoreServ Storage for Client Virtualization and best practices
Figure 4: Scaling HP 3PAR storage from 2,000 users to 4,000 users
Persistent and non-persistent implementations
Numerous use cases are possible in a client virtualization infrastructure. While the details of each use case vary, two basic
types involve: persistent and non-persistent implementations. Each type has different storage requirements:
• Non-persistent VMs—The VM does not maintain any user information directly between sessions; users access any VM
within a pool of defined resources.
• Persistent VMs —VMs maintain pieces of the user’s settings or application stack across reboots, which tie the end user to
a particular virtual machine.
Cost
A client virtualization implementation can range from a small proof of concept design for a small limited number of users to
a large-scale 40,000+-client enterprise environment. Implementing a client virtualization environment offers IT
departments a number of cost benefits.
Centralizing the storage systems drives efficiency. All data is stored in a central location, which streamlines management
and ongoing maintenance.
HP 3PAR StoreServ storage simplifies virtual client administration by reducing storage provisioning and administration time
by up to 90%. HP 3PAR StoreServ storage is autonomic. Autonomic management simplifies, automates, and expedites
storage management in these areas:
• Storage provisioning
• Tiering
• Change management
These changes occur autonomically— intelligently, at a subsystem level, and without administrator intervention.
When used in conjunction with Citrix XenDesktop and VMware Horizon View, HP 3PAR StoreServ storage autonomic
management dramatically simplifies virtual client administration, by reducing storage provisioning and administration time.
Unique HP 3PAR StoreServ built-in features simplify client provisioning and manage the use and deployment of golden
operating system images and OS patch installs.
9
Technical white paper | Why HP 3PAR StoreServ Storage for Client Virtualization and best practices
Setting up and configuring HP 3PAR StoreServ arrays for client
virtualization
One benefit of client virtualization is the ability to rapidly and easily provision new virtual clients. However, achieving a true
end-to-end provisioning solution requires an agile storage infrastructure. Complex storage provisioning and management
are particularly detrimental to changing client virtualization environments. HP 3PAR StoreServ features streamlined
provisioning and management, providing an ideal platform for client virtualization.
Controller nodes
The controller node manages the physical disks and presents them to the host. The HP 3PAR StoreServ controller node is a
proprietary and powerful data movement engine designed for mixed workloads. Controller nodes deliver the high
performance, computing, and connectivity capabilities within the HP 3PAR StoreServ system.
Multiple controller node options are available for a client virtualization environment. Deciding on the controller node options
will vary from one deployment to the next. Using multiple controller nodes enhances connectivity, redundancy, and
performance for your client virtualization environment.
Table 1 shows the total number of controller nodes for each HP 3PAR StoreServ model.
Table 1: Number of controller nodes for HP 3PAR StoreServ models
Model
Number of Controller Nodes
HP 3PAR StoreServ 7000 Series
– 7200
2
– 7400
2 or 4
HP 3PAR StoreServ 10000 Series
– 10400
2 or 4
– 10800
2,4 or 8
Host ports and protocol support
The host port connection allows the array and servers hosting the virtualized clients to communicate with each other. HP
3PAR StoreServ arrays offer both fibre channel and iSCSI protocol host port connectivity options; see Table 2.
One fibre channel advantage over iSCSI is the large number of available fibre channel host ports.
Table 2: Number of fibre channel and iSCSI ports for HP 3PAR StoreServ models
Model
Number of FC Ports
Number of iSCSI Ports
– 7200
4 - 12
0-4—10 Gb/s
– 7400
4 - 24
0-8—10 Gb/s
– 10400
0 - 96 ports
0 - 16 ports—10 Gb/s
– 10800
0 - 192 ports
0 - 32 ports—10 Gb/s
HP 3PAR StoreServ 7000 Series
HP 3PAR StoreServ 10000 Series
10
Technical white paper | Why HP 3PAR StoreServ Storage for Client Virtualization and best practices
Configuring a client virtualization environment varies with environment and budget. Figure 5 shows a configuration using an
HP 3PAR StoreServ 10400 array, HP BladeSystem c-Class enclosure with FlexFabric modules, and HP ProLiant BL460c
server blades using fibre channel connectivity. In this example, 32 fibre channel host ports are used for fibre channel
connectivity. This example shows certain benefits such as multiple workloads spread over multiple devices, failover, and
redundancy.
Figure 5: Example of a client virtualization environment using an HP 3PAR StoreServ array, HP BladeSystem c-Class blade servers, and fibre
channel connectivity
POWER
SUPPLY
POWER
SUPPLY
AC
POWER
SUPPLY
AC
AC INPUT
100-240V~
POWER
SUPPLY
AC
AC INPUT
100-240V~
AC
AC INPUT
100-240V~
AC INPUT
100-240V~
J7
J6
J5
J4
I
CB1
O
POWER
SUPPLY
POWER
SUPPLY
AC
POWER
SUPPLY
AC
POWER
SUPPLY
AC
AC
I
CB0
O
AC INPUT
100-240V~
AC INPUT
100-240V~
AC INPUT
100-240V~
POWER
SUPPLY
AC INPUT
100-240V~
POWER
SUPPLY
AC
POWER
SUPPLY
AC
POWER
SUPPLY
AC
AC
J3
J2
J1
J0
AC INPUT
100-240V~
AC INPUT
100-240V~
AC INPUT
100-240V~
POWER
SUPPLY
AC INPUT
100-240V~
POWER
SUPPLY
AC
POWER
SUPPLY
AC
POWER
SUPPLY
AC
AC
J7
J6
J5
J4
I
CB1
O
AC INPUT
100-240V~
AC INPUT
100-240V~
AC INPUT
100-240V~
AC INPUT
100-240V~
I
CB0
8
7
6
8
7
6
O
J3
J2
J1
J0
0
1
5
RX
RX
FAN
1
FAN
5
1
TX
RX
2
TX
RX
3
TX
4
TX
RX
RX
1
TX
RX
2
TX
RX
3
TX
4
TX
RX
TX
RX
4
RX
TX
3
RX
TX
RX
TX
1
1
TX
RX
2
J7
J6
J5
J4
2
TX
RX
3
TX
4
TX
RX
0
4
3
HALT AND TURN OFF NODE-POWER BEFORE
REMOVING THE NODE OR ADAPTERS.
1
5
4
3
HALT AND TURN OFF NODE-POWER BEFORE
REMOVING THE NODE OR ADAPTERS.
I
CB1
HP 1/10Gb-F VC-Enet Module
1
E1
RX
TX
S0
4
X4
2
3
4
HP 1/10Gb-F VC-Enet Module
UID
1
2
3
4
2
Reset
X5
X1
19
20
21
22
23
0
18
19
20
21
22
23
0
UID
X2
X3
X4
X5
17
18
19
20
21
22
23
0
17
18
19
20
21
22
23
0
4
Reset
8Gb SAN Switch
8Gb SAN Switch
UID
Reset
6
Reset
8Gb SAN Switch
8Gb SAN Switch
RX
3
TX
RX
5
8
RX
2
TX
7
TX
Enclosure Interlink
OA1
1
1
TX
RX
2
TX
RX
3
TX
RX
4
S0
X3
18
17
S1
RX
TX
RX
S1
RX
TX
4
4
RX
RX
TX
TX
3
3
RX
RX
TX
TX
2
2
RX
RX
TX
TX
1
1
X2
17
Reset
UID
O
1
Reset
UID
3
I
TX
UID
X1
E0
2
1
0
2
E0
HALT AND TURN OFF NODE-POWER BEFORE
REMOVING THE NODE OR ADAPTERS.
1
HALT AND TURN OFF NODE-POWER BEFORE
REMOVING THE NODE OR ADAPTERS.
0
3 4 5
E1
3 4 5
O
CB0
OA2
UID
iLO
Reset
Active
Enclosure
UID
Remove management modules before ejecting sleeve
J3
J2
J1
J0
0 1 2
0 1 2
NODE 0
NODE 1
NODE 0
PS
0
PS
1
FAN
6
NODE 1
PS
0
PS
1
NODE 2
PS
0
PS
1
FAN
10
NODE 3
PS
0
PS
1
PS
6
PS
5
PS
4
PS
3
PS
2
PS
1
USB 1
USB 2
HP ProLiant
BL465c G5
HP ProLiant
BL465c G5
1 port 15k
UID
HP ProLiant
BL460c G6
HP ProLiant
BL460c G6
1 port 15k
1 port 15k
UID
serial scsi
UID
serial scsi
146 GB
J7
J6
J5
J4
UID
serial scsi
146 GB
146 GB
HP ProLiant
BL460c G6
1 port 15k
UID
serial scsi
146 GB
HP ProLiant
BL460c G6
1 port 15k
UID
serial scsi
146 GB
HP ProLiant
BL460c G6
1 port 15k
UID
serial scsi
146 GB
1 port 15k
serial scsi
146 GB
NIC
1
NIC
1
FLEX
1
FLEX
1
FLEX
1
FLEX
1
FLEX
1
NIC
2
NIC
2
FLEX
2
FLEX
2
FLEX
2
FLEX
2
FLEX
2
I
CB1
1 port 15k
1 port 15k
O
1 port 15k
serial scsi
serial scsi
146 GB
146 GB
1 port 15k
serial scsi
serial scsi
146 GB
146 GB
1 port 15k
1 port 15k
serial scsi
1 port 15k
serial scsi
serial scsi
146 GB
146 GB
146 GB
I
CB0
O
HP ProLiant
BL460c G6
HP ProLiant
BL460c G6
1 port 15k
UID
serial scsi
HP ProLiant
BL460c G6
1 port 15k
UID
146 GB
serial scsi
1 port 15k
UID
146 GB
serial scsi
HP ProLiant
BL465c G5
UID
146 GB
HP ProLiant
BL465c G5
1 port 15k
UID
serial scsi
146 GB
1 port 15k
serial scsi
HP ProLiant
BL465c G5
UID
146 GB
1 port 15k
serial scsi
HP ProLiant
BL465c G5
UID
1 port 15k
serial scsi
HP ProLiant
BL465c G5
UID
146 GB
146 GB
FLEX
1
FLEX
1
NIC
1
NIC
1
NIC
1
NIC
1
NIC
1
FLEX
2
FLEX
2
NIC
2
NIC
2
NIC
2
NIC
2
NIC
2
1 port 15k
serial scsi
146 GB
1 port 15k
1 port 15k
serial scsi
146 GB
serial scsi
146 GB
1 port 15k
serial scsi
1 port 15k
146 GB
serial scsi
146 GB
1 port 15k
serial scsi
146 GB
FLEX
1
FLEX
2
1 port 15k
1 port 15k
serial scsi
146 GB
serial scsi
146 GB
1 port 15k
serial scsi
146 GB
J3
J2
J1
J0
OK
Drives
Drive technology is a major hardware consideration when setting up and configuring a client virtualization environment.
IOPS, latency, and, drive spin-up times are extremely important. Read/write ratios depend on the users and workloads used.
Solid-state drives (SSDs) handle reads and sequential writes quite well, but offer only marginally better performance over
spinning disks with random writes.
SSDs are more expensive than other drive types.
Before deciding on a drive choice, compare the capacity and speed of the various options available. In some cases, it is
worthwhile to trade off absolute I/O and latency in exchange for growth capability and cost.
Consult with HP when evaluating which drive technologies to use in an HP 3PAR StoreServ storage client virtualization
solution. HP highly recommends using the HP Client Virtualization Analysis and Modeling (CVAM) service for this purpose.
Refer to the HP 3PAR StoreServ QuickSpecs for a list of the supported drives types and sizes.
HP 3PAR StoreServ 7200 and 7400
HP 3PAR StoreServ 10000
11
Technical white paper | Why HP 3PAR StoreServ Storage for Client Virtualization and best practices
Logical components
Chunklets
As shown in Figure 6, a chunklet is a 1 GB block of a physical disk. When a physical disk is admitted to the system, it is
divided into chunklets that become available to the system. Some chunklets are used by logical disks and other chunklets
are designated as spares to hold relocated data during a disk failure or during maintenance procedures. A chunklet is
assigned to only one LD (logical disk).
For more information on chunklets, see the HP 3PAR Architecture white paper.
Figure 6: Chunklet example
LDs and RAID sets for HP 3PAR StoreServ storage
In the context of HP 3PAR StoreServ architecture, LDs are a collection of physical disk chunklets arranged as RAID sets, each
mapped to chunklets from different physical drives. The HP 3PAR StoreServ OS automatically creates LDs with the desired
performance, availability, and size characteristics.
Types of RAID sets:
All the RAID sets described in this section are available for use in your client virtualization environment. Consider the
strengths and weaknesses of each type when deciding which best meets your requirements.
• RAID 0 (block level striping)
RAID 0 is also known as a stripe set or a striped volume. RAID 0 keeps blocks of data in sequence, one disk at a time, for
all disks in a configuration. Because there is no overhead on a RAID 0 set, it is the fastest way to read and write data. Read
and write speeds on the same disk are approximately equal. The main disadvantage of RAID 0 is that there is no parity,
and consequently if a drive fails, all data is lost. RAID 0 offers ideal performance, though it does not offer resiliency.
 RAID 1+0 (mirroring + striping)
RAID 1+ 0 (also known as RAID 10) writes data in stripes across primary disks that have secondary disk mirrors. Client
virtualization works well with this RAID level. Performance increases due to block-level striping, and the replication of
volumes mirrored onto separate physical disks manages additional I/O requests. RAID 10 provides the best overall
benefit for performance and resiliency.
 RAID 5+ 0 (RAID 5 distributed parity +striping)
RAID 5 writes data across a set of hard disks, calculates the data parity, and writes that parity to one hard disk set. RAID 5
then writes the parity to a different disk in the set for every further block of data. Combining RAID 0 striping produces
performance increases. However, RAID 5 has the disadvantage of increasing overall costs and lowering available capacity.
In order to write to a RAID 5 environment, the affected blocks are first read, the changed data is entered, the new parity is
calculated, and the block is then written. On systems with large RAID 5 sets, this means a write I/O is many times slower
than a read I/O, which is undesirable in a client virtualization environment. RAID 50 works best for client virtualization
environments where all data resides on the same logical device. HP 3PAR StoreServ uses an ASIC to accelerate RAID 50
calculations to increase performance.
 RAID 6 (multiple distributed parity, with striping)
RAID 6 is identical to RAID 5 except for the addition of a second parity block. It does not have a performance penalty for
read operations; it does have a performance penalty for write operations, due to the overhead incurred by parity
calculations. RAID 6 offers the same features as RAID 50, with the additional parity block.
12
Technical white paper | Why HP 3PAR StoreServ Storage for Client Virtualization and best practices
LD parameters:
In the HP 3PAR StoreServ architecture, LDs implement RAID functionality. Each LD is mapped onto chunklets to implement
the RAID set. The HP 3PAR StoreServ OS automatically creates LDs with the desired performance, availability, and size
characteristics.
Use the following parameters to control the layout of an LD to achieve specific characteristics. Consult with HP when
deciding how to implement LDs using client virtualization. HP highly recommends using the HP Client Virtualization Analysis
and Modeling (CVAM) service for this purpose.
• Set size
The LD set size is the number of drives that contain redundant data.
• Step size
The step size is the number of contiguous bytes on a single chunklet.
• Row size
The row size determines the level of additional striping across more drives.
• Number of rows
The row size is the depth of the RAID 10 or RAID 50 wide striping.
Common provisioning groups
Common provisioning groups (CPGs) create a virtual pool of LDs that allows Virtual Volumes (VVs), discussed further down
in this paper, to share the CPGs resources and allocates space on demand. CPGs enable fine-grained, shared access to
pooled logical capacity. Instead of pre-dedicating logical disks to volumes, the CPG allows multiple volumes to share the
buffer pool of logical disks. This is extremely valuable for client virtualization because individual CPGs co-exist on the same
array to support multiple client workloads.
As shown in Figure 7, a traditional array requires dedicated drives and dedicated spare disks for each RAID level, while
limiting performance to a single Logical Unit Number (LUN). In contrast, the HP 3PAR StoreServ architecture allows all RAID
sets to reside on the same LDs, using chunklet-based wide striping across physical HDDs, and distributed, rather than
dedicated, sparing. This approach increases overall performance. Compared to traditional arrays that require a dedicated
infrastructure of individual LUNs, the HP 3PAR StoreServ architecture enables the creation of CPGs on a user-defined LD set.
Figure 7: Comparison between CPGs created using HP 3PAR StoreServ LDs and traditional arrays
Best practice
For fastest client response times, use Solid State storage for master images and replicas.
13
Technical white paper | Why HP 3PAR StoreServ Storage for Client Virtualization and best practices
Considerations
Creating multiple CPGs for a client virtualization environment offers several benefits, given varying workloads and user
types. Consider the following items when creating CPGs.
Replicas and master images
A master image is a fully provisioned copy of the operating system shared by all users. For example, a replica is a special
snapshot of a master image in VMware Horizon View environments. The defining characteristic of a replica is a nearly 100%
random read workload. This workload is well suited for VVs created from an SSD CPG providing the fastest disk response
times, to achieve excellent performance for users.
Best practice
Segregating master images and replicas from differencing files and user data optimizes overall performance and cost of
client virtualization storage systems.
Differencing files
Many VDI implementations create a separate file at runtime that records differences between a shared master image and
the end user. The generated I/O patterns are far more random and write intensive than physical environments when the OS
is shared from either memory or a dedicated set of accelerated disks. This workload is well suited for VVs created from
rotating disk media.
User data and settings
Ideally, all users in a CV environment will have their data and user settings redirected away from a central image and onto a
separate CPG. This approach is a major HP 3PAR StoreServ advantage; it allows user data and settings to co-exist on the
same array, without compromising the capability to handle mixed workloads. This workload is well suited for VVs created
from rotating disk media.
Creating a CPG
Use either the HP 3PAR Management Console, shown in Figure 8, or the command line interface (CLI) to create a CPG.
Specify the general characteristics and growth settings for the CPG.
When creating a CPG, plan and monitor the array for unexpected growth. Base the growth settings on the array capacity,
because the array does not allow growth warnings or limits to exceed the amount of currently available storage.
Pay attention to the following key settings:
• Select a name that easily identifies the CPG.
• Select the HDD type and device type that best meets the specified needs.
• Select the appropriate RAID set for the CPG.
Use the Show advanced options checkbox to set warnings and limits, the set size, step size, and other allocation settings.
Best practice
Use names that match the underlying RAID characteristics when creating a CPG.
14
Technical white paper | Why HP 3PAR StoreServ Storage for Client Virtualization and best practices
Figure 8: Creating a CPG using the HP 3PAR Management Console
Virtual volumes and virtual volume LUNs
The HP 3PAR StoreServ architecture introduces the concept of virtual volumes (VVs). A VV is the base, or original volume,
that maps all visible user data to an LD. VVs draw their resources from CPGs. One or more controller nodes may own an LD,
allowing VVs to be striped across multiple controller nodes for additional load balancing, performance, and availability. VVs
store all the data resources for each virtual desktop.
As shown in Figure 9, the host sees only its mapped volume. The volume mapping contains the specifics of the source CPG.
Figure 9: Each host sees only its mapped volume
15
Technical white paper | Why HP 3PAR StoreServ Storage for Client Virtualization and best practices
VV creation
Client virtualized environments often involve multiple storage volumes shared by a cluster of host servers running
numerous VMs. The HP 3PAR StoreServ VV creation process uses the HP 3 PAR Management Console, shown in Figure 10.
The console provides a custom command-line interface for activities such as scripting.
Consider these attributes when creating VVs:
• Volume name: The naming convention identifies the volume and its purpose.
• Allocation setting: Select a thinly or fully provisioned volume. Fully provisioned volumes have a set amount of user space
allocated to user data storage. Thinly provisioned VVs obtain space from the CPG as needed and allocate space on
demand in small increments. Each method has distinct benefits. Consult HP before deciding which allocation method to
implement. The “Thin versus full allocations” section of this document includes details on thinly provisioned versus fully
provisioned volumes.
Use the HP 3PAR Management Console to set warnings and limits for thinly provisioned volumes.
• Size: Select a size that will meet the volume requirements.
• User CPG: Select the CPG that fits the desired usage model. The CPG contains all the attributes used by the VV.
Figure 10: VV creation using the HP 3PAR Management console
16
Technical white paper | Why HP 3PAR StoreServ Storage for Client Virtualization and best practices
Exporting a VV
Virtual Volumes are only visible to a host as virtual LUNs (vLUNs), which are exported in the following ways:
The HP 3PAR Management Console in Figure 11 shows the method of exporting a VV.
 To specific hosts (set of worldwide names or WWNs)
Use this method when the VV is visible to the specified WWNs, regardless of the port on which they appear. This is a
convenient way to export VVs to known hosts.
 To any host on a specific port
Use this method when the hosts (or their WWNs) are not known before export, or in situations where the WWN of a host
cannot be trusted (host WWNs spoofing).
 To specific hosts on a specific port
VVs do not consume LUN numbers as they do on some systems (only vLUNs consume LUN numbers).
Figure 11: Exporting a VV using the HP 3PAR Management Console
Thin versus full allocations
Client virtualization storage requirements vary dramatically as hundreds to thousands of diverse clients boot up, operate,
and power down. The HP 3PAR StoreServ family offers two distinct storage allocation methods: fully provisioned and thinly
provisioned volumes. Each method offers its own unique benefits.
Full provisioning
Fully provisioned volumes provide physically mapped storage provisioned directly for a specific host. These volumes map
directly from the LD to the virtual client through VVs. Fully provisioned volumes are efficient when the amount of resources
used and the amount of resources allocated is about the same. They make the most sense for storage needs that are
already defined.
Thin provisioning
Thin provisioning provides as much capacity as needed over the lifetime of all virtual clients, while only allocating the
physical capacity actually consumed. Allocated storage is presented to host systems using thin-provisioned VVs (TPVVs).
TPVVs are mapped to a logical CPG, which serves as the common storage reservoir (unlike VVs, which are pre-mapped to
17
Technical white paper | Why HP 3PAR StoreServ Storage for Client Virtualization and best practices
underlying LDs and ultimately to chunklets). When writes are made to the TPVV, the CPG creates the mapping to underlying
LDs and allocates space.
A client virtualization deployment generating write intensive workloads to TPVVs can enhance the VV performance using the
saga command option during SSD CPG creation. Refer to Appendix C: Configuring HP 3PAR StoreServ Snapshot
Administrator (SA) Space for details.
Figure 12 shows a thinly provisioned volume allocated to a set size. The additional headroom is useful for situations where
extra space is needed to support future growth and to reduce the risk of application failures. However, if this headroom is
never required, it will result in wasted physical HDDs and unwarranted expense.
Figure 12 also shows how thin provisioning provides higher storage use by eliminating the need for additional HDDs.
Storage is used as the environment grows, delivering increased performance and lower overall hardware cost.
Figure 12: Thin versus full provisioned volumes
Best practice
Thin Provisioning reduces wasted storage by allocating only the physical capacity consumed when it is needed.
HP 3PAR StoreServ Flat SAN and Direct Attached Storage
In client virtualization environments, up to 80% of network traffic is server-to-server and server-to-storage. Legacy, multitier, fibre-channel network architectures require a complex web of network cards, interconnects, cables, and switches to
keep pace with this changing traffic. This approach creates performance bottlenecks and drives higher costs and complexity
to build, maintain, and secure at scale.
To address this problem HP announced HP Virtual Connect for 3PAR with FLAT SAN technology, the industry's first direct
connection to fibre channel-based storage that does not require dedicated switches. This technology provides a simplified
architecture with best-in-class storage and network innovations such as BladeSystem, Virtual Connect, and the HP 3PAR
StoreServ.
Customer benefits include these features:
• Single layer fibre channel storage network eliminates SAN switches and Host Bus Adapters (HBAs)
– Massive simplification of fabric management
– 2.5x faster fibre channel storage provisioning
– 50% less infrastructure costs
– Automated zoning capabilities to set up zones with one-click
– Up to 55% less latency by removing the SAN fabric layer
18
Technical white paper | Why HP 3PAR StoreServ Storage for Client Virtualization and best practices
Figure 13: HP Flat SAN Architecture
SAN versus DAS
The question of whether to use a Storage Area Network (SAN) or Direct Attached Storage (DAS) for the core differential files
in a client virtualization environment generally revolves around what level of persistence those files must have. For
example, in an environment where all users connect to non-persistent VMs, DAS may be a better choice. HP IO Accelerators
or industry leading Smart Array controllers connected to either internal or external storage that scales at the disk level may
be a cost -effective way to service end user needs while providing a compelling end user experience. If the need to maintain
VM mobility across hosts exists, a SAN should become a point of discussion. Similarly, an all–persistent environment
introduces the need for SAN as a critical checkbox.
In environments using both persistent and non-persistent VMs, IT administrators and decision makers need to ask the
following:
• Do I want to manage more than one approach to storing data?
• With non-persistent VMs, is my requirement to migrate VMs for purposes of system maintenance fixed or can I simply
rely on controlling user access to free up systems?
Answers to these questions will vary by company, but in all instances, a SAN is required.
The Virtual Connect Flat SAN supports only HP 3PAR StoreServ storage. For more information on using Virtual Connect Flat
SAN with HP 3PAR StoreServ storage, refer to this website:
Refer to the Virtual Connect Flat SAN site for more information
3PAR file services and Network Attached Storage (NAS)
A successful client virtualization implementation requires the segmentation of user data from the images used in CV
implementations. This segmentation adds to the robust nature of CV and allows users to move from one VM to the next
with their data following them as they go. HP 3PAR File Services provides a front end for user data access while aggregating
the data on a highly available and resilient HP 3PAR StoreServ back end. Based on HP StoreEasy 3830 Gateway Storage,
3PAR File Services offers these advantages:
• Network File System (NFS)
• Common Internet File Services (CIFS)
• Object Storage
• Inline thin conversion and autonomic thin reclamation
• Deduplication with 50 to 60% space savings
• Microsoft clustering for High Availability
• Volumes of up to 16TB
19
Technical white paper | Why HP 3PAR StoreServ Storage for Client Virtualization and best practices
The StoreEasy 3830 is itself a robust platform that offers resiliency with these features:
• Health monitoring
• Advanced memory protection
• Redundant fans and power supplies
• Mirrored OS disks
• Clustering
StoreEasy is tightly integrated with 3PAR arrays. Administrators can provision LUNs on 3PAR arrays from within the
StoreEasy interface and manage groups of file servers from one place, rather than opening multiple interfaces. Along with
the standard management interfaces and available plugins for common virtualization management consoles, this allows
administrators in CV scenarios a choice of how to work that maximizes their efficiency and achieves maximum visibility.
StoreEasy also allows for the management of data based on its business value, helping fulfill one of the promises of CV. IT
administrators may manage the NAS, but they have little insight into which files are business critical or may contain
proprietary information. With StoreEasy, content owners who know the data can categorize it themselves. Administrators
then use centrally managed policies to provide that data protection. Within a CV scenario, these features mean that
sensitive, proprietary data is safe.
File Services for 3PAR are continuously available in a persistent VM environment. In client virtualization implementations,
capabilities for transparent failover, online patching, and file system availability are critical. In many instances, the core
virtual machine that the user connects to is disposable, but in a persistent VM environment, everything is backed up and
user settings are maintained. Full-site failover operations are possible with HP Cluster Extension (CLX); overall solution
availability returns in seconds to minutes, instead of hours.
Disaster Recovery: Peer Persistence
HP 3PAR StoreServ storage delivers the industry’s best total high availability to minimize downtime during catastrophic
failures and ensure that users have uninterrupted productivity, even in the most mission-critical client virtualization
environments. HP 3PAR Peer Persistence software enables HP 3PAR StoreServ systems located at metropolitan distances
to act as peers to each other, presenting a nearly continuous storage system to hosts and servers connected to them. This
capability allows administrators to configure a high-availability solution between two sites or data centers where failover
and failback remains completely transparent to the hosts and applications running on those hosts. Another advantage of
3PAR Peer Persistence is that it enables load balancing between sites. As a result, administrators can move hosts from one
site to another based on business and performance needs, without affecting the applications running on those hosts. Figure
14 provides details on Peer Persistence for VMware environments.
Figure 14: Peer persistence using HP 3PAR StoreServ Storage and VMware
20
Technical white paper | Why HP 3PAR StoreServ Storage for Client Virtualization and best practices
Best practices for optimization
HP 3PAR StoreServ Adaptive Optimization (DO) and Dynamic Optimization (AO) software are optional features that allow
storage administrators to seamlessly migrate data between storage tiers in their environment without interrupting data
access.
HP 3PAR Dynamic Optimization Software
Storage Administrators use HP 3PAR Dynamic Optimization Software to move volumes to different RAID levels, drive types,
or both, and to redistribute volumes after adding drives to an array.
CPGs define RAID level and disk type. DO alters the VV service level by moving the VV to a different CPG. In HP 3PAR
StoreServ, DO provides the flexibility to place data on CPGs that meet current service level requirements, without binding
this data permanently to that configuration. This proves powerful for client virtualization because currently used volumes
are moved to other CPGs. Administrators adjust the workloads for those volumes to different RAID sets and HDD
technologies to meet the workload demand.
Figure 15 shows how to use DO to choose the service levels of a previously created volume, changing RAID set sizes and
drive type. Administrators can make changes while normal processing is underway without affecting the volume, and can
alter the service levels in these cases:
• Incorrect performance information has been supplied
• Volume is either under- or over-utilizing provisioned resources.
Figure 15: Dynamic Optimization can be used to change the service levels of a previously created volume
Best practices for Dynamic Optimization and client virtualization
DO is desirable in the following cases:
Volume layout changes after hardware upgrades— As a client virtualization environment grows, the administrator is able to
increase the existing VVs without having to delete them and recreate new ones. Existing VVs only take advantage of
resources present when a volume is created. If a storage server is upgraded by adding cages or disks, the original volume
layout may no longer be the most effective one. DO allows administrators to change the VV layout to take advantage of the
new hardware.
Volume RAID level changes—Administrators are able to convert volumes from one RAID set to another as requirements
change, because different RAID sets have varying capacity requirements and offer differing degrees of performance. For
example, if a pre-existing environment is using RAID 50, the set can be changed to RAID 10.
Volume availability level changes—The availability of a virtual volume determines its level of fault tolerance. For example, a
volume with cage-level availability can tolerate the failure of a drive cage because its RAID sets use chunklets from different
drive cages. Administrators may alter the availability characteristics of existing VVs as application and business
requirements change within the client virtualization environment.
Volume service level changes—Administrators may change the volume parameters (such as disk filtering parameters)
applied when a volume was created.
21
Technical white paper | Why HP 3PAR StoreServ Storage for Client Virtualization and best practices
Peak I/O consideration
During peak I/O demands where SSDs are limited in a configuration, consider using DO to move volumes from lower
performance drives to SSDs. With peak I/O workloads such as login and boot storms, using DO to move these workload
intensive volumes around as needed will maintain the overall performance of your client virtualization environment.
HP 3PAR Adaptive Optimization Software
HP 3PAR Adaptive Optimization (AO) Software is a granular, policy-driven, autonomic storage tiering solution that delivers
service level optimization for virtual and cloud data centers. With AO Software, granular (sub-volume) data movement takes
place autonomically based on user-configurable analysis periods, collected performance data and built-in stability
mechanisms.
Best practices for AO in a client virtualization environment are beyond the scope of this paper. Please note that CV
workloads are significantly different than typical enterprise application workloads and AO must be carefully configured to
meet the specific needs of this workload. The incorrect usage or configuration could result in sub-optimal performance
results. Please contact your HP local systems engineer for additional information on the appropriate technologies and
information for tiering in client virtualization environments.
Client virtualization products
Exclusive virtualization and automation features built into HP 3PAR StoreServ storage work and integrate with third party
products to deliver unique benefits and increase virtualization ROI.
HP 3PAR StoreServ storage supports the following virtualization products:
• VMware vSphere
• VMware Horizon View
• Microsoft Windows Server Hyper-V
• Citrix XenServer
• Citrix XenDesktop and XenApp
Summary
Client virtualization poses a multitude of challenges for IT infrastructure. Before choosing any technology, it is strongly
recommended that a complete and thorough analysis of the current desktop environment be conducted. To assist you with
this, HP provides a Client Virtualization Analysis and Modeling Service (CVAM).
To ensure the best possible ROI, it may be desirable to co-host client and server virtualization infrastructures, as well as
infrastructure for other mission critical applications on the same systems. HP 3PAR StoreServ storage delivers built-in
mixed workload support to deliver high ROI.
Performance is a paramount concern when deploying storage to support client virtualization. HP 3PAR StoreServ storage
has several built-in features to ensure the highest possible performance for your client virtualization project. Because of the
I/O blender effect, client virtualization can pose significant challenges for any store system. Deploying a storage platform
that can support the extreme I/O requirements of boot and login storms while cost effectively supporting the requirements
of daily client logins is critical.
Capacity optimization, resilience, and support for sizing and scaling are additional critical requirements for storage that
supports client virtualization. HP 3PAR StoreServ delivers industry-leading unique features in each of these areas.
These are among the unique advantages that make HP 3PAR StoreServ storage the ideal foundation for building or
expanding a client virtualization environment to meet the needs of the Instant-on enterprise.
For more information on HP 3PAR StoreServ storage for client virtualization, please visit hp.com/go/cv or
hp.com/go/storeserv.
22
Technical white paper | Why HP 3PAR StoreServ Storage for Client Virtualization and best practices
Appendix A: Host Persona ALUA Support for ESX
Host personas are a set of behaviors that allow the host to deviate from the default host behavior and use primitives
specific to the OS. By assigning a persona to a host, multiple host types that require distinct customized responses can
share a single host port. For example, hosts running Windows and VMware operating systems can all connect to the same
host port. This simplifies connecting hosts to the array and reduces management costs related to complex host
connections.
HP 3 PAR OS 3.1.2 supports the Asymmetric Logical Unit Access (ALUA) command for ESX hosts. Figure A-1 shows the
process of selecting the host persona ALUA using the HP 3PAR Management Console.
Figure A-1: Host Persona ALUA Support for ESX
23
Technical white paper | Why HP 3PAR StoreServ Storage for Client Virtualization and best practices
Appendix B: Effect of read caching on storage performance
Optimized implementations of client virtualization tend to use a shared image concept. For example, with VMware View this
can appear in one of two forms.
• The first is through a linked clone/replica model where the replica (a special snapshot of the master image) is hosted on
SSD or other accelerated media, resulting in an outstanding end user experience and also offloading OS and application
specific reads in any given desktop workload. This model alleviates and accelerates all reads regardless of current state.
• VMware Host Caching uses a small amount of memory (up to 2GB with View 5.01) to cache common OS information. The
effect is the same with nearly all workload reads removed from the SAN layer. Citrix Provisioning Server (PVS), part of the
XenDesktop suite, also uses a memory caching approach, though a larger functional cache is standard.
Theoretical discussion
Desktop workloads vary widely from implementation to implementation. For example, workloads between 10% and 90%
reads and those ranging from 3 to more than 70 IOPS can occur. The primary benefits of caching a variety of workloads are
twofold:
• The normalization of I/O patterns to the rotating media, whether SAN or DAS
• The reduction in overall I/O through the removal of OS and locally installed application reads to accelerated media
(memory or flash based disk).
The second effect is obvious. The amount of the overall reduction is directly related to the size of the cache and the amount
of read I/O generated within the application and OS stack. The normalization may not be as intuitive, but should become
obvious with a brief discussion.
Assume a workload that is straightforward at 50% reads, 50% writes, and a total of 20 IOPS per user. Were this image
deployed on a desktop, or even as a fully provisioned VM, the underlying disks could be expected, on average, to have to
serve up 1 read for every 1 write and approximately 20 I/Os per second. If a storage administrator were to plan a SAN for
100 users at this workload, the SAN would need to support (ignoring peak workloads), a steady stream of 2,000 IOPS at a
50/50 read/write ratio.
Shifting this model to a shared image or cached model (whether that means caching in memory or locating replicas on
SSD/flash media) immediately alters the workload characteristics to rotating disk. Assume that all of the I/O generated is
coming from either an application or OS image. In reality, some I/O will come from end user data, but that data should be
relocated to a file share and planned separately. With 50% of workload as reads, all of which come from an OS or
application, workload is immediately altered by caching off those reads. Instead of 20 IOPS, of which 10 are reads and 10
are writes, 10 reads are lost. What is left is 10 writes to the disk. However, this is not entirely useful for planning. As the
differential disk (whether linked clone, write cache file, or differencing disk) ages, it will build up data that may itself need to
be read from rotating media. To help account for this, HP has adopted a strategy of adding back reads as 10% of remaining
writes when planning the rotating disk layer. In this case, this means adding 1 read to the 10 writes. A storage administrator
can now plan the SAN for 11 IOPS per user at a roughly 90% write workload rate.
The conclusion to draw is that the use of optimized read caching features within common client virtualization product suites,
such as VMware View (Linked Clone/Replica and host caching) and Citrix XenDesktop (PVS), can dramatically reduce the
amount of I/O to the SAN, but it also transforms the workload in the process.
24
Technical white paper | Why HP 3PAR StoreServ Storage for Client Virtualization and best practices
Appendix C: Configuring HP 3PAR StoreServ Snapshot Administrator (SA)
space
This section explains how to configure a CPG to allocate SA (Snapshot Administration) metadata for Thin Provisioned
Volumes from a different tier; in the following example the CPG is configured to use the FC tier for user data while the
metadata (SA) will be configured to reside on the SSD tier. This should be a best practice for heavy write workloads when
SSDs are available in the systems.
1.
2.
The array should have SSD drives installed.
The TPVVs should be created from an FC CPG (referenced as FC_CPG in this document). Create the CPG with the
following CLI command:
createcpg –saga “-p –devtype SSD” <CV_FC_R10>
This command will allocate all SA space associated with the CPG used to create the TPVVs on available SSD drives. This
command does not specify a CPG as the target of the SA space, just a drive type. The CPG specified in the command
(FC_CPG) will allocate its SA space from SSDs.
Acronym list
• Common Provisioning Groups (CPG)
• Computer Internet File System (CIFS)
• Direct Attached Storage (DAS)
• Hard Disk Drives (HDDs)
• Host Bus Adapters (HBAs)
• Logical Drives (LDs)
• Logical Unit Number (LUN)
• Network File System (NFS)
• Operating System (OS)
• Physical Drives (PDs)
• Redundant Array of Independent Disks (RAID)
• Solid State Drives (SSDs)
• Storage Area Network (SAN)
• Virtual Desktop (VD)
• Virtual Desktop Infrastructure (VDI)
• Virtual LUNs (vluns)
• Virtual Volumes (VVs)
25
Technical white paper | Why HP 3PAR StoreServ Storage for Client Virtualization and best practices
For more information
HP Storage Arrays
HP Storage, hp.com/go/storage
HP 3PAR Storage Arrays
HP 3PAR StoreServ Storage Family, hp.com/go/storeserv
HP 3PAR StoreServ Storage Best Practices Guide,
http://h20195.www2.hp.com/V2/GetDocument.aspx?docname=4AA4-4524ENW
HP 3PAR StoreServ Get Virtual Guarantee Program,
http://www8.hp.com/us/en/products/data-storage/data-storage-products.html?compURI=1284456
HP 3PAR Adaptive Optimization Software Solution Brief,
http://h20195.www2.hp.com/V2/GetDocument.aspx?docname=4AA3-3176ENW
HP 3PAR System Reporter Reference User Guide,
http://h20000.www2.hp.com/bizsupport/TechSupport/DocumentIndex.jsp?contentType=SupportManual&lang=en&cc=us&
docIndexId=64179&taskId=135&prodTypeId=18964&prodSeriesId=5044528
HP 3PAR StoreServ Implementation Guides,
http://h20000.www2.hp.com/bizsupport/TechSupport/DocumentIndex.jsp?lang=en&cc=us&contentType=SupportManual&
prodTypeId=18964&prodSeriesId=5044394&docIndexId=64179#2
HP Servers
HP Servers, hp.com/go/servers
HP Support for VMware
HP Unified Sizer for Server Virtualization, hp.com/solutions/sizer/unified-sizer-server-virtulization
HP Insight Control Storage Module for VMware vCenter Technical Documentation,
http://h18004.www1.hp.com/products/servers/management/unified/infolibraryicv.html
HP Insight Control Storage Module for VMware vCenter Download,
https://h20392.www2.hp.com/portal/swdepot/displayProductInfo.do?productNumber=HPVPR
HP Support for Client Virtualization
HP Support for Client Virtualization, hp.com/go/cv
To help us improve our documents, please provide feedback at hp.com/solutions/feedback.
Sign up for updates
hp.com/go/getupdated
© Copyright 2013 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.
4AA4-6479ENW, April 2013