VDI Solutions Guide. Solving VDI Challenges with Affordable 100% Flash Storage

VDI Solutions Guide

Table of Contents

2

Introduction to 100% Flash VDI

3

The Evolving VDI Landscape

- The Toxic Nature of VDI on Traditional Disk Storage - The Five Key VDI Storage Blockers

- Different Approaches to Next-Generation VDI Storage

11

The Pure Storage All-Flash Solution for VDI

- Solving All-Flash Economics with Inline Data Reduction - The All-Flash User Experience

- The All-Flash Admin Experience

- Dramatically Simpler Storage Administration - Pure Storage VDI Reference Solutions - Comparing to Typical Disk Solutions

20

Conclusion

21

Case Studies

- City of Davenport: VDI & Server Consolidation in One Array - STI: Enabling Bring Your Own Device with Flash

Introduction to 100% Flash VDI

The IT industry has been abuzz over the past several years promoting the idea of VDI: virtualizing and centralizing desktops to enable IT to deliver a more secure, manageable, less costly, and ultimately more mature end-user computing model. While the dream of pervasive deployment of virtual desktop infrastructure has been discussed and tried for decades, the recent explosion of x86 virtualization, and availability of commodity scalable server architectures with increasingly larger amounts of CPU power and centralized memory have brought the promise of VDI much closer to reality. In fact, IT departments are finding that with the right investments in infrastructure, VDI can indeed deliver a client computing model that is both better for the end-user (a truly mobile, multi-device computing experience with better performance than dedicated devices) and better for the IT staff (centralized management, consistent security and policy enforcement, resiliency through device independence, and enablement of “bring your own device” BYOD models).

So if VDI comes with so many potential advantages, why has adoption of VDI been so slow (roughly 1.5% of enterprise desktops are virtualized today)? The reality is that the path to achieving the VDI promise land is a difficult one, and many organizations have abandoned their VDI initiatives outright or left them in partial stages of deployments. The reasons why are many, but most failed deployments boil-down to three key issues:

• Too expensive: VDI is often justified as a technology initiative to reduce desktop cost, but in reality most find that they are unable to achieve the promised ROI due to infrastructure costs. In particular server, networking, and storage devices can be dramatically more expensive than dedicated desktops/laptops.

• Poor end-user experience: if VDI isn’t implemented properly, the end result is slow or unavailable desktops that lead to user frustration and lost productivity.

• Too difficult to manage: VDI shifts the desktop administration burden from the end-users to IT staff. While this affords many security and administrative benefits, it also means more work for often burdened IT staff,

especially if the VDI environment isn’t architected correctly.

More often than not, the root cause of all three of these failure modes is storage. Traditional disk-based storage is optimized for high-capacity, modest performance, and read-heavy workloads – the exact opposite of VDI which is write-heavy, very high performance, and low-capacity. The result is that as performance lags, spindle after spindle of legacy disk storage has to be thrown at VDI, causing a spike in infrastructure costs and a spike in management complexity.

In this paper we’re going to explore how a new, 100%-flash based approach to VDI can overcome the key VDI failure traps, and deliver a VDI solution that both end-users and IT administrators will love. We’ll examine in-depth why flash storage is the right choice for VDI, and how Pure Storage helps lower the cost of all-flash storage to allow VDI to be affordable. Finally, we’ll introduce a pre-tested and validated Pure Storage Reference Architecture for VDI, which provides a variety of configurations for scaling your VDI environment from 100s to 1,000s of users seamlessly (more detail on the reference architecture can be obtained in a stand-alone document available at purestorage.com/VDI).

The Evolving VDI Landscape

VDI Finally Goes Mainstream

Despite years of fits and starts in the VDI market, VDI technology and IT process maturity around VDI deployments have matured to the point where VDI has finally “crossed the chasm” to adoption by mainstream enterprises. In fact, a recent survey at the Gartner Data Center Conference showed >50% of attendees were planning to embark on desktop virtualization projects:

Alongside this move to mainstream adoption, how people plan, justify, and deploy VDI solutions has changed. In this section we’ll explore the shift to next-generation VDI, and how that impacts storage and broader infrastructure choices.

VDI 1.0

!

VDI 2.0

The first generation of VDI deployments were largely focused on call centers and other similar task workers, with the goal of locking down the environment with a consistently configured stateless desktop. In these niche use cases VDI provided a clear benefit in terms of reduced administrative burden and increased security, and the demands of this style of user on the infrastructure were modest at best (<5 IOPS/desktop was a common load on storage). In this 1.0 generation the main driver was cost: virtual desktops proved to be cheaper desktops for this use case, and much of the cost savings came through stateless desktops (VMware View Linked Clones or Citrix Provisioning Server Images), which use the virtualization layer to deduplicate images, only storing a master image and personalization/configuration information to reduce storage spend.

What Applications Are You Planning to Virtualize?

2011 End-User Survey at Gartner Data Center Conference

>50% of

respondents

plan desktop

virtualization

projects in

the next year

!

!

The usage model around VDI is changing, however, and a new second generation of VDI deployments is emerging, with different target users and different requirements. The second generation of VDI is being implemented for productivity users and executives, users who typically have a higher performance need and expectation of user experience. The drive towards second-generation VDI is also different: the mobile workforce is increasingly demanding flexibility in their compute infrastructure, wanting to have a consistent computing experience independent of location (home vs. office, local vs. traveling) and of access device (desktop, laptop, tablet, even smart phone). IT sees this second-generation of VDI as an opportunity to bring more users under the VDI umbrella, offer them better and more reliable computing, and enable new models like BYOD (“bring your own device”) for further flexibility.

It turns out that this second generation of VDI has major differences in terms of VDI project justification and funding. VDI 1.0 was ultimately defined by cost: task users were viewed as “cost center” consumers of computing, and as such the bar for justifying a VDI 1.0 deployment was ensuring that it has positive ROI, and was ultimately less expensive than physical desktops. In

second-generation VDI, the target user base is a higher-value user, and as such the IT

organization is often willing to spend more to deliver higher-quality computing, and both productivity and uptime of that user is of paramount importance.

The VDI 2.0 model also has significant differences in terms of the infrastructure required to support it. Cost is still important, and cost remains the #1 challenge that disrupts VDI deployments of all shapes and sizes, but a new set of requirements enter as we move to VDI 2.0. The biggest difference is that the 1.0 model relied heavily on stateless desktops,

VDI 1.0

!

VDI 2.0

!

task

workers

!

desktop clients

!

!

$

!

$

$

!

!

cost

!

!

!

!

$

!

$

$

!

!

stateless

(linked clones)

!

!

Stateless Desktops (Linked Clones)

Non-persistent desktops that are created by the virtualization hypervisor & connection broker, based upon a master “gold” image. Each desktop is created fresh at login, but does not persistently store any user data, changes, or configuration. Ideal for call center or task worker use cases.

Persistent (Stateful) Desktops

These VDI images are complete desktop images where all customization and state are retained between sessions. As a result, Persistent Desktops consume dramatically more storage space per desktop, but are more flexible. Ideal for productivity user, executive, or mobile user use cases.

while the 2.0 model relies much more on persistent desktops. Users want access to their desktop from these different devices, and not saving or caring about the user’s desktop information and state is simply no longer an option. Once desktops become stateful and large percentages of the organization utilize the VDI infrastructure, availability becomes a chief concern – VDI becomes mission-critical and you simply can’t afford to have user downtime. Finally, the infrastructure must scale to new levels of both performance and capacity – most VDI deployments already see performance challenges at 100s of users, and in the 2.0 model are expanded to 1,000s of users, and often hit the breaking point.

As organizations evolve along the VDI maturity path, what they come to realize is that “one size fits all” simply doesn’t work to deliver VDI 2.0. In the first generation VDI efficiency was gained by making all VDI images the same, but as the VDI deployment expands it becomes clear that not all corporate users are the same, and most organizations will end-up implementing a variety of different user profiles, some using stateless desktops and some using stateful desktops, with different resource consumption and performance profiles. Net net – the VDI infrastructure needs to evolve with VDI to be able to meet the varied requirements of VDI 2.0.

The Toxic Nature of VDI on Traditional Disk Storage

difficult workload VDI is for traditional spinning-disk based enterprise storage. Traditional storage arrays are optimized for typical enterprise applications, which tend to be read-IO focused, have dedicated LUNs for each application, have relatively constant and predictable workloads, and rely on caching from DRAM and/or flash to get performance when disk falls down. The problem is that VDI is almost the 180-degree opposite of what traditional storage is optimized for: it is write-heavy (80% writes, 20% reads), highly random in its access patterns, and often requires up to 20 IOPS/desktop, which can equate to 100,000s of IOPS in a typical deployment. And to make matters worse, there are periods throughout the day (login storms,

patch/maintenance cycles, and virus scans) where the workload spikes or changes completely.

The net result is that traditional storage arrays simply can’t keep-up with most VDI deployments, and customers then go to extreme methods to try and patch the IO problem with work-arounds. These work-arounds come in one of three flavors: buy more disk spindles to get more IOPS, buy more arrays to get more IOPS, or add a flash cache to get more IOPS. The first two work-arounds lead directly to bloated spending on storage, and the administrative complexity and inefficiency that comes with using way too much disk. And unfortunately, the flash caching solution rarely works: VDI is write-heavy, and most flash caching solutions only accelerate reads.

!

!

10Ks to 100Ks of IOPS

Highly Random

80% Write, 20% Read

Bursty & Unpredictable

The Five Key Storage VDI Blockers

Because of the challenges highlighted above, storage is most often the core culprit in failed or de-scoped VDI deployments. We’ve identified five key “storage VDI blockers” – ways that traditional storage can derail your VDI deployment:

1: High storage cost kills VDI ROI

The ROI behind VDI deployments is fragile at best: replacing a bunch of commodity desktops with datacenter server/storage gear isn’t cheaper, so you have to believe in the broader efficiency, productivity, and security benefits to make the case. But if you peel back the cover on the ROI case, you’ll likely find that the #1 cost contributor to VDI deployments is storage. Storage starts expensive, and when you find that you can’t meet your performance or scale objectives, expanding storage gets

even more expensive. It’s not uncommon to see storage be 60% of the infrastructure cost of VDI, and $/desktop costs of storage range from $500s - $1000 / desktop. If you see your VDI costs skyrocketing, chances are storage costs are to blame.

2: Failure to scale from pilot

!

roll-out

Most VDI deployments start with a small pilot, a couple hundred users that are typically!the IT staff. Things seem to go well on traditional storage, and the environment is likely only generating a few thousand IOPS. The organization decides to move forward, and scale from 100s of users to 1,000s – and this is where the trouble starts. The storage array that everyone thought could handle 1,000s of users simple fails under the load of all those VDI sessions, and to make matters worse it can’t be

expanded – there is a hard limit to the amount of IO that the storage array can handle. So the only choice becomes either a) buy another storage array, or b) scale-back the VDI deployment. More often than not this is where the budget reaches its end, and option B is the only option.

3: Poor end-user performance

Most end-users initially fear VDI – there is a sense of loss of control, and ultimately being at the mercy of the IT department to do the basic computing required to get their jobs done. As such, when users transition to the VDI infrastructure they expect it to be bad, and their sensitivity to poor performance is unnaturally high (they tend to

immediately forget the 10-minute boot time their older Windows XP laptop was experiencing). The other thing to understand about VDI is that it is user interactive – every user action ultimately is fed back to the servers which transact IO to the storage, and that storage IO is in the latency chain that translates directly to user experience.

Key to selling VDI to end users is delivering a better user experience than their dedicated computers, which is absolutely possible given the high performance of the servers and storage they are now using in the datacenter – but only if the infrastructure is designed with user experience in mind.

As you can probably guess, blockers 1, 2, and 3 are inherently connected and form a chain reaction in many VDI deployments: the VDI infrastructure is scaled, leading to a break-down in storage performance. Users start to experience poor performance and availability, leading to calls to the help desk and complaints. IT looks to beef-up the storage infrastructure to improve the experience only to see the cost of another storage array – and suddenly the executive staff is questioning the ROI case for the entire VDI deployment.

4: Stateless desktop

!

stateful desktop transition balloons storage usage

An often undiscussed blocker is the storage capacity impact that comes with moving from stateless to stateful desktops. Many organizations start their VDI deployment assuming a rigid uniformity in using a single image and stateless desktops. But as they get into the project (and as they move from VDI 1.0 to VDI 2.0) they realize that many if not a majority of their desktops need to be stateful. And with stateful

desktops come storage problems: much more storage space consumption, and higher requirements for resiliency and multi-site access. Stateful desktops are just one more challenge that creates an often-unexpected balloon in storage spending, and undermines VDI ROI.

5: Extreme difficulty in storage administration

Finally, in addition to being overly expensive and hurting the VDI ROI, traditional storage is also simply too complex for most VDI administrators to deal with. VDI is often managed by the end-user computing team not the storage or server admin teams, and as such they are not nor want to be storage experts. Enterprise storage starts off difficult to begin with, but the added complexity required to fine-tune performance and expand storage when VDI

performance demands it often proves to be too much pain for VDI admins to deal with. As such, traditional storage not only ends up sapping a large portion of the VDI budget, but is also a leading time sink for VDI admins, leading to bottom line operational cost challenges as well.

Different Approaches to Next-Generation VDI Storage

It has become clear to everyone that a different approach to storage is needed to enable VDI to cost-effectively scale across the enterprise. Four main approaches have been suggested to solve the VDI storage challenge, and we’ll look at the benefits and issues with each in this section.

Host-Based Flash Caching

VDI vendors such as VMware with View have enabled features to allow the virtualization infrastructure to take advantage of host-resident flash in the form of a PCIe SSD or DAS SATA/SAS SSD in the VDI server itself. While these host-based SSD solutions help to accelerate read IO and can add performance to stateless desktop environments by essentially fully caching the golden master image, they do nothing to help with write IO, which turns out to be the bulk of VDI IO. The caches are also not highly-available, are only used by a single VDI server so must be purchased for every server, and add to the VDI administrative burden by introducing a second tier of storage. Net net: host-based caching will remove some VDI bottlenecks, but isn’t the most efficient or useful way to adopt flash for VDI.

Hybrid Arrays: Array-Based Flash Caching / Tiering

Most major disk-based storage arrays have moved to adopt flash caching and/or tiering as a way to introduce a small amount of flash into the disk array to accelerate IO operations (a typical deployment is 5% flash and 95% disk). The size, granularity and mode (read-only vs. read & write acceleration) of the flash caching varies, but in all cases the arrays use cache algorithms to try and watch and/or predict which blocks are most heavily accessed on the array and move or copy those into the flash cache to get better performance from those blocks. While flash caching offers some modest performance enhancement for VDI, the reality is that a small amount of flash simply can’t make up for the large amount of slow spinning disk in traditional arrays. VDI is a write-heavy application, so while a read cache will help a bit with login storms, it won’t help accelerate most operations. Few of the solutions offer write caches, and if they do ultimately a write cache is limited in size and as it fills it becomes limited by the back-end disk’s ability to drain data from the cache down to disk. Net net: flash caching is a modest band-aid on traditional disk-based storage, but won’t add the performance necessary for most VDI deployments.

!

!

Converged Compute / Storage Appliances for VDI

A wide range of converged compute/storage appliances have entered the market, some focused specifically on the VDI use case. These appliances range from high-end solutions which integrate multiple vendor’s gear into a “pre-configured rack” to lower-end solutions which integrate servers and DAS storage into a single appliance. All these appliances use a combination of disk and flash storage to accelerate VDI performance. The main focus for these appliances is not performance per se, but rather focusing on reducing the administration burden of VDI. Because these appliances typically employ a small amount of flash and a large amount of disk, all the arguments above around hybrid arrays apply equally to these solutions. It is important to remember that what the user gains in configuration ease, they lose in flexibility in converged solutions. VDI is a complex application, where the balance between compute, memory, storage performance, and storage capacity all have to be tuned carefully for efficiency. In traditional deployments the virtualization layer makes it straightforward to adjust or scale any of these dimensions independently, but with converged appliances running out of any of these resources means buying another appliance. Also, converged appliances result in the infrastructure being 100%-dedicated to VDI, which is fine for a larger deployment, but adds considerable overhead for smaller customers who may want to share infrastructure between VDI and other applications. Net net: converged appliances don’t help VDI performance

HOST-BASED FLASH CACHING

ADVANTAGES DISADVANTAGES

+ accelerates boot time for stateless desktops - doesn’t help persistent desktops as much + helps recompose times for stateless desktops - only caches read IO, not write IO

- must be installed in every VDI server - not highly available

- another tier of storage to manage

HYBRID ARRAYS: ARRAY-BASED FLASH CACHING / TIERING

ADVANTAGES DISADVANTAGES

+ is shared by entire VDI environment - read cache is of limited utility for write-heavy VDI + works for stateless and stateful desktops - write cache typically small and unable to cache

entire VDI write load

- overall array limited by the performance of disk - lack of predictability in end-user experience

much, and trade-off administrative simplicity for expandability and flexibility down the road. They also guarantee that VDI becomes an island, separate from the server and storage administration and standardization efforts.

!

!

100% Flash Storage for VDI

The final approach to solving the storage-for-VDI problem is to use 100% flash storage for VDI. Most people agree that if you can afford it, 100% flash would be the ideal storage for VDI. 100,000s of IOPS at the ready, ultra-low <1ms latency, and predictable performance can deliver an end-user computing environment that consistently out-performs dedicated desktops, and speeds VDI administrative tasks. Unfortunately – until now all-flash storage has been prohibitively expensive, simply pricing itself out of the VDI use case for most customers. Net net: all-flash storage is the best alternative for delivering the best VDI experience, but going all-flash has been prohibitively expensive.

As you’ve probably guessed by now, Pure Storage’s mission is to make 100% flash storage affordable, and we’ve found that we can do so in particular for VDI workloads. Read on to understand Pure’s reference architecture for affordable 100%-flash VDI.

!

CONVERGED COMPUTE / STORAGE APPLIANCES FOR VDI

ADVANTAGES DISADVANTAGES

+ simplicity – one HW platform for storage & compute - rely on caching for consistent performance + often management integration with VM stack - unable to scale CPU, memory, and storage

independently – run out of any and buy another - limited storage footprint – not expandable - limited performance – CPUs shared for everything

100% FLASH STORAGE FOR VDI

ADVANTAGES DISADVANTAGES

+ the best VDI end-user experience - too expensive for mainstream deployment, breaking the + the best VDI administrator experience VDI ROI case (until now!)

+ accelerates both stateless and stateful desktops + accelerates both read and write VDI operations alike + advantages in power, size, and cooling

The Pure Storage All-Flash Solution for VDI

Introducing Pure Storage

Pure storage was founded with a simple goal in mind: that 100% flash storage should be made affordable, so that the vast majority of enterprise applications can take advantage of the potential advances that flash memory delivers. As such we designed our core product, the Pure Storage FlashArray, from the ground-up for the unique characteristics of flash memory.

The FlashArray’s entire architecture was designed to reduce the cost of 100% flash storage, and it combines the power of consumer-grade MLC flash memory with inline data reduction technologies (deduplication, compression, thin provisioning) to drive the cost of 100% flash storage to be inline or under the cost of traditional enterprise disk storage. Data reduction technologies are particularly effective in VDI environments, typically providing >5-to-1 reduction for stateless desktops and >10-to-1 reduction for stateful desktops.

It’s important to note that unlike some flash appliances, the FlashArray was designed with enterprise-class scale and resiliency in mind. That means a true active/active controller architecture, online capacity expansion, and online non-disruptive code upgrades. The FlashArray also employs a unique form of RAID protection, called RAID-3D™, which is designed to protect against the three failure modes of flash: device failure, bit errors, and

performance variability.

Last but not least, the FlashArray is the simplest enterprise storage that you’ll ever use. We’ve designed from the start to remove the layers of complexity of LUN, storage virtualization, RAID, and caching management common in traditional arrays, and have integrated management directly into VMware vSphere’s Web Client, making

management within a VDI environment seamless.

100% MLC Flash

High-Performance

Inline Data Reduction

always deduped, compressed, thin and encrypted

Resiliency & Scale

high availability snapshots RAID-3D™ online expansion

Solving All-Flash Economics: Inline Data Reduction

Traditional VDI reference architectures rely on exclusive use of stateless desktops to keep the storage usage in-check. Pure Storage implements inline data reduction technologies (both deduplication and compression), which perform inline reduction without impacting overall performance. The FlashArray’s inline data reduction is truly unique: competitive hybrid arrays typically either don’t implement data reduction, or do so only on the flash tier. Competitive flash appliances either don’t implement data reduction, do so at a high performance cost to the user, or don’t implement both deduplication and compression. It turns out that both deduplication and compression have a tremendous cost and performance benefits for VDI.

The clearest benefit that inline data reduction provides is that it levels the playing field between stateless and stateful desktops: both perform the same on flash, and both deduplicate down to taking similar (negligible)

amounts of space on the array. Let’s look at an example: we recently created 2,000 VDI images on the FlashArray with VMware View, 1,000 were stateless (linked clones), and 1,000 are stateful (persistent desktops) that were 31GB in size. The following GUI screenshot shows the storage consumption on the array after provisioning and extensive use of the VDI images:

Had this been a traditional storage array, the 1,000 31GB stateful images would have taken 31TB of disk space on their own. On the FlashArray the VDI images were reduced to taking only 935 GB of space on the array, and with an additional 288 GB of space consumed for RAID protection, the total space used on the array was less than 1.2TB for all 2,000 images. This was the by-product of more than 25-to-1 data reduction.

It should be noted that this represents an idealized environment, as the virtual desktops are “fresh” and identical and no amount of user interaction and customization has happened. As users begin to use the desktops and personal setting are applied, it is common to see the data reduction reduce to roughly 5-to-1 for linked clones, and 10-to-1 or greater for persistent desktops. Furthermore, if user data is stored within the desktops data reduction numbers will be slightly lower, but most user data also reduces quite well on average.

As you can see, Pure Storage allows any combination of stateless and stateful desktops to be deployed, and dramatically reduces the storage space consumption for VDI.

The All-Flash User Experience

End-users demand the best response times from VDI – in fact they are perhaps unfairly unforgiving of anything perceived as latency from the VDI infrastructure. The challenge is that the VDI workload varies wildly throughout the day from the morning boot to afternoon productivity time to evening maintenance. Pure Storage provides ample performance – 100,000s of IOPS that keep VDI running smoothly throughout the VDI day:

Let’s look closer at what happens during a boot storm and login storm on the Pure Storage FlashArray. Boot storms cause large spikes in read IO, often challenging traditional disk arrays and reducing service times to an untenable state for all users during the boot and login period. The following graph shows the IO load on the FlashArray while both booting and logging-in to 1,000 virtual desktops. The total boot time took less than 15 minutes, and the IO load was as follows:

IOPS

!

Morning Noon Evening Overnight Boot

Storm

A / V Scans _Maintenance & Patches

Boot Storm

• Boot 1,000s of desktops in mins • 100Ks of IOPS at the ready • Booting users don’t impact

running users Login Storm • Read-intensive mornings • 20x improvement in login times Predictable User Experience • Consistent <1ms response times • No caching variability Virus Scanning

• Heavy read period • Speed scan times • Lower impact

Patching & Recomposing

• Heavy write period • Recompose time

reduced from hours to minutes

The 1,000 desktops are booted and logged-in to in less than 15 minutes, and as you can see, the FlashArray easily handled the boot load with latency never peaking above 1.1ms, and IOPS staying well within the operating range of the FlashArray (up to 200,000 IOPS).

Pure Storage recently conducted a series of tests on VMware View infrastructure with the View Planner tool. View Planner simulates large-scale VDI deployments by spawning large numbers of VMs (1,000 each of linked clones and persistent desktops or full clones in this case), and executing common user tasks to simulate a busy workday. A “passing” score is to complete all tests with <1.5 seconds response time, as you can see below the Pure Storage FlashArray dramatically exceeded these results, with an overall View Planners score of 0.52 seconds.

VMware View Planner Test – 1,000 Linked Desktops and 1,000 Persistent (Full Clone) Desktops

Ultimately the View Planner tool helps validate that the end-user experience of running on a scaled VMware View infrastructure, and validates that the FlashArray is capable of delivering a best-in-class VDI end user experience, even when hosting 1,000s of VDI seats.

Pure Storage View Planner Score:

0.52 seconds

1,000 Linked Clone Mean Response Time

1,000 Persistent (Full Clone) Mean Response Time Test Pass = under 1.5 seconds

The 100% Flash Admin Experience

We’ve been discussing the advantages that all-flash VDI provides for the end-user, but it turns out that the advantages for the VDI administrator are just as dramatic. While it is true that VDI provides a much more mature administration model for desktops, one shouldn’t underestimate the resulting burden in that administration: desktop administration and maintenance now falls squarely on the shoulders of the VDI admin team, and at times that burden can be large.

We tested a wide variety of administrative tasks that comprise the VDI day, looking at both task completion time, as well as the IOPS load imparted on the FlashArray. Below are some of the results:

Administration tasks on the FlashArray typically took anywhere from 5-10x less time to complete. These savings add up when an administrator executes 10s to 100s of these tasks daily!

Provision 50GB Desktop From Template

Disk

5 min

50 sec

Boot 100 VDI desktops

Disk 15 min 22 sec Recompose 100 VMs Disk 45 min 3 min Storage vMotion 50GB VM Disk 4 min 25 sec THE ADMIN ADVANTAGE OF ALL-FLASH STORAGE IS SIMPLE: What if every administrative action you took completed 5-10x faster? How much more productive would you be?

Dramatically Simpler Storage Administration

The verdict is in: VDI administration is already hard enough; VDI admins don’t want to moonlight as storage admins as well. Unfortunately, deploying VDI on traditional storage practically requires a PhD in storage administration: WWNs, LUNs, RAID types, caching, tiering, spares, utilization levels…enough! One just has to take one look at the configuration diagram from a market leading array vendor’s VDI reference architecture below to see how difficult deploying VDI reliably on a hybrid disk/flash array can be:

Pure Storage was built from the ground-up to be dramatically simpler: the array presents itself as one large pool of storage that can be presented to hosts as LUNs. No worrying about RAID levels – everything is protected system-wide with zero management RAID-3D. No worrying about performance tuning, system-wide striping, flash caching, flash tiering, cache hit rates….every IO is flash-fast, all the time.

Whether your VDI infrastructure choice is Citrix XenDesktop or VMware View, the hypervisor of choice for most deployments these days is VMware vSphere. For VDI environments built on vSphere, Pure Storage has invested in additional management integration to make the FlashArray a seamless extension of your vSphere management workflow. First, the entire Pure Storage GUI is available within the vSphere Web Client, enabling full management of Pure Storage from within vSphere. Second, we’ve created a Pure Storage vSphere Datastore Viewer, which allows customers to drill-down into specific datastores to get complete datastore-to-LUN mapping, and to understand detailed datastore-level capacity, data reduction, and performance information – all without ever having to look up a LUN serial number or switch between admin consoles. A screenshot of the datastore viewer is shown below:

Major Competitor VDI Reference Architecture*

!

VDI on Pure Storage

!

VDI

Infrastructure User Data Other Data

Finally, new datastore creation is a process that is typically complex today, requiring multiple steps on both the storage array and within vSphere. Pure Storage ships a set of datastore creation and maintenance dialogues, which are one-page wizards that allow the fully-automated creation, grow/shrink operations, and deletion of vSphere datastores. An example of the datastore creation wizard is shown to the right; enter the datastore name, size, format, and which hosts/clusters to create it in, and the

FlashArray automates the rest of the process (creating the Pure Storage Volume, defining the hosts, exporting the LUN to the right hosts, rescanning the hosts for the LUN, and creating the datastore).

Pure Storage VDI Solutions

One of the consistent challenges in managing VDI deployments is managing scale throughout the VDI pilot !

production lifecycle. Customers often want to start small to pilot test VDI with a limited set of users, and then want to incrementally grow their deployments as their VDI deployments mature. All too often these deployments start on traditional disk arrays which can handle the performance needs of a pilot, but hit an IO performance wall as the project moves into production – causing either abandonment or costly additional storage array purchases (often array performance can’t be expanded, requiring the purchase of additional arrays, not just additional storage). Pure Storage has done extensive testing of VDI environments in both our lab and jointly with customers, and has created a set of validated reference configurations for VDI deployment. These configurations are designed to start small, and grow with your deployment:

VDI pilots can start with a special ½-shelf non-HA FlashArray configuration, and as the project goes live, a second controller can be added to the array to ensure high availability. As the project grows, additional SSDs and eventually additional storage shelves can be added to the array to meet additional user image and data storage needs. Our smallest offering scales seamlessly (and without downtime) into our largest offering, allowing customers to scale from 100s to 5,000+ VDI seats in a single FlashArray.

!

Pilot

100s-1,000

2.75 TB

10-20 TB

Go Live

Up to 1,000

2.75 TB

10-20 TB

Expand

Up to 2,000

5.5 TB

20-50 TB

Scale-Up

5,000+

11 TB

50-100 TB

Stage

Users

Raw Capacity

Usable VDI Capacity*

Much more detailed configuration, performance, and validation information on the Pure Storage VDI Reference Architecture for VDI can be found in a stand-alone technical document, available at:

Comparing to Typical Hybrid Reference Architectures

As mentioned above, traditional storage arrays are not particularly well suited to meeting the performance needs of VDI deployments economically. To get a sense for the inefficiency, one only has to read the VDI reference architecture from the market-leading traditional array vendors. We surveyed the publicly available VDI reference architectures for 1,000-2,000 seats, and normalized them all to 1,000-seat deployments, averaging the results. On average, these reference architectures suggest using approximately 26TB of spinning Tier 1 disk, and a few hundred GB of flash as a cache. When pricing these arrays in High Availability deployments with typical software attached, average pricing is $200K - $300K, which yields a $200-$300/desktop cost for just the storage. It is also important to understand that all of these reference architectures require the use of stateless images (linked clones) to make the solutions work. If users were to deploy stateful desktops on these solution blueprints, many more TBs of disk would be required, and the economics would worsen still.

Compare the above typical configuration with Pure Storage: the FlashArray can service 1,000 users with our smallest storage configuration, and enables the user to deploy any combination of stateless or stateful desktops. This configuration is also easily expanded by simply adding storage to the configuration.

!

Competitor TBs Disk Spindles GB Flash

D 43 48 600 N1 14 24 256 N2 20 34 205 E1 28 63 0 E2 23 35 250 Average* 26 TB 41 262 GB

* Most recent competitor VDI 1,000-2,000 user reference architectures, normalized to 1,000 users as of 10/2012

Average Disk Array VDI Deployment

•  26TB of 15K disk, 260GB flash

•  $200,000 - $300,000 in cost

($200-$300 per desktop)

•  <50,000 IOPS potential

•  Relies on Linked Clones

•  Scaling-up difficult/expensive

With Pure Storage:

•  2.75 TB of Flash

•  <$100,000 in cost, <$100/desktop

•  Up to 200K IOPS, performance room to spare

•  Use Persistent Desktops or Linked Clones

Conclusions

As organizations move to the next generation of VDI deployments, continuing to rely on traditional spinning-disk storage isn’t a viable option to keep-up with the performance, scale, and reliability requirements of VDI. You simply can’t risk your VDI deployment on the poor end-user performance or poor ROI that is inherent in disk-based VDI implementations.

The Pure Storage FlashArray’s unique architecture which combines 100% flash memory with inline data reduction enables 100% flash VDI to be affordable, resulting in four key advantages:

The FlashArray’s performance and economics have been validated through extensive lab and customer testing, and a set of detailed Pure Storage reference architectures for VDI are available at www.purestorage.com/VDI.

!

!

A

ff

ordable:

100% flash for

<$100/desktop

The All-Flash

User Experience

Scale &

Reliability

Simplest

Administration

Deliver to your customers a better end-user experience: VDI security and mobility with better performance than a laptop with local SSD.

Traditional enterprise storage can cost >$300/desktop. Pure Storage enables all-flash VDI for <$100/ desktop, persistent or stateless. With your organization’s productivity hanging in the balance, VDI is now a mission-critical application. Pure Storage delivers full active/active resiliency with no single point of failure. VDI admins don’t want to be storage

admins. Pure Storage is fully-managed within VMware vSphere, and is the easiest storage array you’ll ever manage.

Now that Pure Storage has broken the price barrier for VDI on

100% flash storage, why risk your VDI deployment on disk?

DATA REDUCTION

DATA REDUCTION

BEFORE 2.1TB

AFTER

Achieved 8-to-1 data reduction with hundreds of persistent virtual desktops Consistent <1ms read and write latency with sustained workload

Planning to move as many as 2,000 virtual desktops to the FlashArray

Migrated 2TB virtual desk-tops on VMware View 5.1 to the Pure Storage FlashArray FA-320 248 GB 8-to-1 PERFORMANCE IMPROVEMENT PERFORMANCE IMPROVEMENT SOLUTION SOLUTION CIO, STI Erik VanLaecken

Pure Storage gave us the performance we needed to make VDI a reality.

The FlashArray is now being deployed in the new school year, where it will be used in the initial pilot deployments of virtual machines for a small set of programs. As they see success with the program, they plan to expand the program across multiple program areas on campus, which could be thousands of virtual desktops.

The Results:

Southeast Technical Institute chose Pure Storage because of its ability to handle the random and unpredictable I/O stream. With hundreds of thousands of IOPS and the need to meet the “instantaneous” response times expected by students, STI discovered the performance of the FlashArray by executing a series of benchmarks against the array.

The FlashArray showed that it could handle an IOPS load as much as 6X higher than the traditional disk array. While the previous disk array delivered 20ms latencies for 50 desktops, the FlashArray delivered sub-millisecond latencies across 100+ virtual desktops running on VMware View’s recent 5.1 release. When combined with a 8-to-1 deduplication ratio, the FlashArray delivered the performance at a cost the institute could afford. This outcome enabled the IT department to bring the first departments online with their VDI vision for Southeast Technical Institute.

The Answer:

Southeast Technical Institute is South Dakota’s largest post-secondary technical institute. They specialize in educating students in 50 different career fields, so they can join the workforce confident in their chosen profession. Many of their programs require practical training with expensive, curriculum-specific software.

To support their programs, the IT department issues as many as 2000 laptops pre-loaded with the needed applications. At the end of a semester, the IT department would have to secure all the laptops from the students, then re-image them or upgrade to newer versions of the software. An additional challenge was the online student community, who could only get access to the software by paying expensive licensing fees.

The IT department at STI saw what desktop virtualization could do to improve their operational efficiency and agility, but they also recognized that their existing

storage infrastructure was too slow to handle the large and random workload that so many virtual machines would produce. To solve their performance problem, STI still needed to stay within their fixed IT budget.

“We looked at many mainstream storage solutions, including the traditional mechanical disk arrays as well as hybrids, but none of them could match Pure Storage for overall performance, space and consistency. Pure Storage gave us the performance we needed to make our vision for a VDI-enabled student community possible. Now, we are more responsive to our clients, can upgrade software packages in the middle of a semester, and will enable the campus to begin moving towards a BYOD environment,” said Erik VanLaecken, CIO at STI.

The Challenge: 2000 Laptops with Proprietary Software

FLASHARRAY DELIVERS THE COST AND PERFORMANCE TO MAKE IT A REALITY

Southeast Technical Institute’s VDI Vision

BEFORE 10 TB

AFTER 2.7 TB

Initiated pilot VDI deployment using a FlashArray and vSphere ESX 5.0 and used linked clones

Reduced average write latency from 20ms to 1ms Reduced average read latency from 17ms to 0.5ms

Linked clones replication from 45 to 2 minutes Sufficient performance to mix workloads on the FlashArray 6.5-to-1 DATA REDUCTION DATA REDUCTION PERFORMANCE IMPROVEMENT PERFORMANCE IMPROVEMENT SOLUTION SOLUTION Network Administrator, City of Davenport Cory Smith

Anyone who is going to do a VMWare View deployment should put their linked clones on a Pure Storage array. It’s essential.

The City of Davenport IT infrastructure deployment project is progressing with the storage performance they need at a budget they can support. The City of Davenport plans to increase their VDI deployment up to 500 or more! Because departments have decided to use virtual desktops instead of buying physical desktops, the FlashArray has already paid for itself within their budget. And there’s still extra capacity to take on additional workloads.

Without the FlashArray, the successful deployment and maintenance of these hundreds of VMs using linked clones in VMWare View would not have been possible within the budget, even with mechanical disk.

The Results: VDI Outperforms Physical Laptops

City of Davenport decided to trial VMWare View and a Pure Storage FlashArray concurrently. They deployed a pilot of VDI users, starting with 150 desktops. Since then, department heads have started calling IT asking when they can have all their physical desktops replaced with VMs. Why? Because with the FlashArray, the virtual machines are faster than the physical desktop machines.

The environment comprises six vSphere(tm) ESX 5.0 hosts and the workload is a 70/30 mix of reads/writes. The applications running within these virtual machines are Windows-based, with Exchange 2007 and SQL dominant in the workload. Test results on these workloads were tremendous! “When we move people back to the mechanical array, we get a lot of complaints,” said Cory. That’s because with the FlashArray, Davenport IT was able to reduce latency to less than 1ms, down from 20ms.

City of Davenport now uses linked clones to improve maintenance and upgrade efficiency of virtual desktops. By creating clones of a VM, installing and upgrading user applications can be done on a single VM instead of across many. However, using linked clones concentrates IO to a specific location in a mechanical array, which slows all machines down. Using the FlashArray, a linked clone environment is easily handled because the Purity Operating Environment disperses the IO and data across the entire array.

The Answer: Replace Mechanical Disk

with a Pure Storage FlashArray

Like many municipalities, the City of Davenport wanted to transition to the more flexible and efficient IT infrastructure afforded by virtual desktops (VDI). The City of Davenport IT department supports hundreds of employees, while needing to provide a specific level of service to those employees to ensure their productivity.

However, the mechanical disk-based array they were using wasn’t able to meet the performance requirements for their initial VDI pilot deployment of 50 VMs; the randomization of the IO stream caused by virtualization slowed the existing array beyond what would allow the IT team to succeed in getting new users on board. “Had we tried to force users to move to VDI using the disk array, the project would have failed,” said Cory Smith, City of Davenport IT.

The Challenge: HDD-based Array Can’t Support VDI

VDI DEPLOYMENT SUCCESS WITH LOW LATENCY PURE STORAGE FLASHARRAY

City of Davenport, Iowa Conquers VDI Performance

!

Pure Storage, Inc. Twitter: @purestorage 650 Castro Street, Suite #400 Mountain View, CA 94041 T: 650-290-6088 F: 650-625-9667 Sales: [email protected] Support: [email protected] Media: [email protected] General: [email protected]