SlideShare a Scribd company logo
1 of 6
Download to read offline
Summary
This paper reports on a configuration forVirtual Desktops (VDs) which reduces the total
hardware cost to approximately $32.41 per desktop, including the storage infrastructure.This
number is achieved using a configuration with dual node, cross-mirrored, High Availability
storage. In comparison to previously published reports, which tout the storage infrastructure
costs alone ofVDI at from fifty to several hundred dollars per virtual machine, the significance of
the data becomes self evident. In this report, storage hardware costs become inconsequential.
Even more significantly, the reported configuration achieves this result with 220VDs running
on a simple pair of low cost servers.The primary innovation consists in co-locating the
redundant storage virtualization systems onto the same hardware platforms as theVDs,
thus eliminating the need to amortize high-cost, stand alone storage controllers against
manyVirtual Server platforms and thousands ofVDs. Previous publications have reported
on configurations which use thousands ofVDs to defray the cost of these controllers.
Reading between the lines, it becomes immediately apparent that perVD hardware costs
rise very sharply as such configurations are scaled downward.Yet, it is precisely these smaller
VDI configurations which are the more important from most practical standpoints.
On the other hand, this configuration may also be scaled upwards, in a linear
fashion, to thousands ofVirtual Desktops, thus eliminating distended configurations
created by the search for artificial “sweet spots” at which costs are optimized.
Finally, this configuration uses theVSI benchmark and is based on DataCore’s SANmelody
software and the Microsoft Hyper-V virtualization platform.The use of Hyper-V in such
benchmarking is unusual as theVMware ESX platform is typically used forVDI sizing. DataCore
expects similar results with ESX and will publish those results when they become available.
Benchmarking a Scalable and
Highly Available Architecture
forVirtual Desktops
by Ziya Aral & Jonathan Ely - DataCore Software
WHITE PAPER
The Problem
Modern Hypervisors use Storage Area
Networks (SANs) to provide the storage
infrastructure for networks of virtual
machines and hosts. Such shared
infrastructures are used to provide portability,
provisioning capability, and flexibility to
VMs. Moreover, the need for such networks
grows proportionally as the numbers of
VMs are multiplied. As the number of eggs
in “the basket” grows, so too does the need
for High Availability configurations to keep
them running and to prevent outages which
encompass, not one or two, but perhaps
dozens of machines. AsVMs proliferate,
routine management tasks, such as backup
and migration, increasingly depend on the
storage management functionality of SANs.
Multiply these requirements by a factor
of 10 or 100, and we arrive at the synergy
betweenVirtual Desktops and SANs.
The problem forVirtual Desktop
implementations is that SANs are often
implemented with large and costly storage
controllers and complex external storage
networks. While these have the advantage
of achieving reasonable scalability, they
introduce a very large threshold cost toVDI
implementations.To overcome this burden,
hardware vendors typically benchmark
with one to several thousandVDs.
Virtual Desktops are at their introductory
stage. Many companies, while understanding
the potential benefits of the technology, are
introducing pilot programs or attempting
to fit initialVDI implementations into
existing structures. If the granularity ofVDI
implementations is to be in the thousands,
then the user is forced to consume, not just
the “whole loaf”, but an entire bakery truck
full of loaves at one sitting … and this before
even knowing whether the bread tastes good.
The alternative is equally unappetizing.The
user “bites the bullet”, and accepts the high
threshold cost and complexity of a full-blown
SAN while running far less than the optimal
number ofVDs. Now, the per-desktop cost
of the implementation becomes much larger
than it would have been if the “old scheme”
of discrete desktops had remained.This is
quite an introduction to a new, “cost-saving”
technology… as an increasing number of
ever-practical bloggers have noted.
DataCore
This benchmark is run using the DataCore’s
software-based storage virtualization
environment on the same platform as the
hypervisors hosting theVirtual Desktops.
The DataCore software is a full featured
implementation of a High-Availability SAN,
with all of the functions necessary for aVM
hosting infrastructure and unusually high
performance characteristics. In addition to
that, the software has several characteristics
which recommend it for this application.
Because the software is portable, it can run
as aVM under Hyper-V or ESX and may also
run on Windows Server 2008 at OS-level,
alongside Hyper-V. While this means that
it is a consumer of a common hardware
resource, it also means that many of the
most important SAN interconnects and
storage structures (such as block cache)
are implemented locally and virtually – on
top of a common backplane. Such a
scheme thus gains more from its locality
to its clients than it “costs” by virtue of
being a consumer for common hardware
resources, as will be elaborated in the
Performance Discussion section below,
In addition, the scheme described simplifies
the overall configuration by scaling
transparently – each time a newVDI host
is added, so are the additional storage
resources needed to feed it. Moreover, the
scaling occurs in both directions, eliminating
what is otherwise a difficult problem in
appropriately sizing infrastructure.
None of this should be interpreted as a
“tossing of the gauntlet” by software based
controllers against hardware ones. DataCore
often runs in conjunction with such hardware
controllers. If the user has a need for such
devices, then there is no fault. But, if there
is no absolute requirement for such devices
at the foundation ofVDI architectures – if
neither cost nor capability argues for their
initial inclusion – then, hardware controllers
serve only to skew theVDI architecture.
The Benchmark
The benchmark used is theVirtual
Session Indexer (VSI) Pro 2.1 from
Login Consultants.VSI is becoming the
standard for this type of workload.
2
In comparison to the “roll-your-own”
benchmarks used by several of the storage
vendors, and, even in comparison to trace
data,VSI appears to be just a shade more I/O
heavy (and especially WRITE-weighted).
VSI also creates a working set which
is somewhat larger than might be
expected from discrete desktops.The
explanation for this is thatVSI runs
every element of its emulated desktop
environment on every individualVD. In
addition, application synchronization
acrossVDs is expressly minimized.
On the whole, this is easy to justify
as simple workload normalization,
necessary for the creation of a repeatable
benchmarking environment.VSI, though,
errs on the conservative side and is a bit more
challenging than many proprietary workloads.
Configuration
The configuration used in this benchmark
consists of two “white box” server nodes,
each with an ASUS Z8PE-D18 Dual
LGA motherboard, 2 Intel Xeon E5640
Westmere 2.66GHz Quad Core CPUs and
an 80GB SATA boot drive.The total cost
is $2114.14 per server, exclusive of DRAM
and application storage. In addition, each
node is configured with 64GB of DDR3
PC3-10600 DRAM (at a cost of $1056.24) and
4 application drives: 2 SAMSUNG HD103SJ
SATA disk drives and 2 Western Digital
VelociRaptor WD3000HLFS SATA disk drives
($393.60 for all 4).Total server hardware
costs are thus $3564.68 per machine and
$7,129.36 for the full configuration.
Two SATA drives are configured as a Storage
Pool, belonging to SANmelody. SANmelody
runs at OS level in this configuration.
Whether SANmelody runs at this level,
or as aVirtual Machine itself, makes
little difference. In this case, the former
was chosen for benchmark visibility.
The Storage Pool, once configured, is
used to host a “golden image” of theVDs.
It is then “snapped” using the DataCore
Snapshot facility, and each of the writable
“snaps” is presented as a boot LUN to the
VDs. Experiments were tried to determine
if multiple golden images might improve
performance. Not only did they not help
3
but the existence of multiple source
images actually increased the amount of
memory required to accommodate the
working set in DataCore’s block cache.
Finally, the original image and its snapshots
are mirrored to the second server (using
DataCore synchronous mirrors over iSCSI)
to achieve High Availability. Not only is the
configuration thus proof against storage
failure but theVDs may be restarted on
any other server if a more catastrophic
failure should occur.Two SATA drives are
configured as a second storage pool to
receive mirror traffic from the other node.
The configuration described above
is implemented with a few simple
scripts which then launch 110Virtual
Desktops per server and hand the entire
configuration off to theVSI benchmark.
We live in an era in which the primary
characteristics of computing machinery
of any class are etched in silicon and are
thus nearly identical for any member of
that class.This configuration was chosen
in order to create a baseline reference at
the “lowest common denominator”.The
choice of server platforms and their costs
will vary by the substitution of everything
from specific brand names to optimizations
for footprint, power consumption, and
certain enterprise “features”. Still, none of
these variations will impact performance
significantly, and may be seen as simple cost
deltas to the reference presented here.
Results
The configuration below was able to
host 220 Desktops using theVSI 2.0.1
benchmark.The disclosure and benchmark
configuration report is reproduced in
full as an addendum to this paper.
Performance Discussion:
Co-residency –The principle innovation
in this benchmark result is the use of
DataCore’s storage virtualization system
on the same hardware platforms used to
hostVirtual Desktops. While conventional
wisdom might suggest that the DataCore
software would thus become a competitor
for the same scarce resources of the
hardware platform, such as memory and
CPU, numerous experiments proved the
opposite to be true. In each case, such co-
located configurations easily outperformed
configurations with external storage.
The reasons are easy to isolate,
retrospectively.TheVirtual Desktop
application is not particularly I/O intensive
and proves to be easily served by DataCore
at the cost of very few CPU cycles.The
elimination of the lion’s share of external
channel traffic serves to further reduce
those demands. In return, block-cache
latencies become nearly nonexistent, channel
overheads disappear, and I/O latencies
are eliminated.The resident SANmelody
“pays” for itself, without losing anything
in the way of capability or portability.
From the standpoint of memory, a similar
dynamic also holds.Virtual Desktops
provide for relatively small working sets
which tend to be READ heavy. Since all
READs go to the single source volume,
which in turn is the basis for the differential
system volume for eachVirtual Desktop,
the entire grouping caches very well
in a relatively small cache size. Larger
caches produce diminishing returns.
Finally, one predictable advantage to
these types of configurations is that they
are inherently “self-tuning”. Each time a
group ofVirtual Desktops are added, so too
is the storage infrastructure necessary to
support them. In contrast to generic SANs
which are open ended in their requirements
and often organized as independent, and
sometimes complex, storage networks, in
this case the known requirements of the
VDI application may be used to make the
storage infrastructure largely transparent.
Disks – Given efficient WRITE caching,
required disk performance may be
significantly minimized. In this benchmark,
4 SATA drives were used, configured into
two separate DataCore 2-disk pools.The
first pool was used for primary storage
and the second was used for mirroring the
alternate node.The use of a mix of 7,200
and 10,000 rpm devices is not significant.
Any four (and perhaps even 3) 7,200 rpm
SATA devices of similar characteristics to
the Samsungs produced similar results.
The importance of this distinction for us
is contained in a separate discussion on
creating a baseline configuration. We were
loath to use anything but the most standard,
easily configured SATA components. SAS,
Fibre Channel spindles, fast devices, hybrid
disks, and SSDs all have their place in the
real world and are well understood to deliver
a multiple of ordinary SATA performance.
Still, the incorporation of such devices in
baseline benchmarking has the same effect
as buildingVDI architectures around this
or that storage array. Such optimizations
may become the dozen different tails
attempting to wag an increasingly confused
dog. Not only do they make comparisons
difficult, but they lead to the subtle intrusion
of a specific hardware architecture into
the realms in which theVDI requirements
themselves should be paramount.
4
Memory – The cost of DRAM comprises
a significant portion of the total platform
costs in this configuration. Decreasing the
size of the memory for eachVD is thus
a natural temptation for optimizing the
number ofVDs supported by each platform.
The difficulty is that the one constant in
our industry is the continuously changing
density and cost structure of DRAM... a fact
which argues for the avoidance of DRAM
optimization, rather than arguing for it.
This difficulty is compounded by the
fact that both Microsoft andVMware
offer hypervisor features which allow the
“over-loading” of the memory allocated
to individual virtual machines. In effect, a
kind of global virtual memory is created
across “machines”, though the actual
implementation mechanisms may differ.
Finally, it was discovered during the course
of DataCore’sVSI benchmarking thatVDs
with very small memory allocations – so small
that they would have prevented a discrete
PC from booting - nevertheless ran normally
in this configuration. On examination, it
appeared that the performance of the
DataCore block cache created a paging
hierarchy which did nothing to prevent
“thrashing” but did practically absorb it when
a match with working set sizes was achieved.
In the end, neither “overloading” nor “fast
thrashing” were used in the implementation
described here. Not only did this allow
for fair comparisons to benchmark data
which had already been published, but it
avoided optimization for schemes which
are largely proprietary and independent
of theVDI sizing exercise. In a word,
the conventional allocation scheme
used here avoided “diversions”.
The actual DataCore cache size employed
was 4 GBs. While larger caches produced
somewhat better results, this was the point
at which “diminishing returns” could be
clearly established. Subtracting this usage
and the memory allocation for Windows
Server 2008, the remaining DRAM was
simply divided by the number of resident
VDs in a conventional fixed allocation.
Software Costs – It might be protested that
since DataCore has used a software-based
StorageVirtualization system in place of
a hardware controller, that the cost of the
software must be reported in order to create
a true “apples-to-apples” comparison.This
brings with it its own challenges. Most of
the hardware controllers which have been
benchmarked are fairly minimalist. Much
of the feature set bundled into SANmelody
is either not available or it is available at
additional cost to the reported configurations.
With that qualification noted, the total
cost of the software was $3848 per node or
$34.98 perVirtual Desktop.This includes
both the cost of Microsoft Server 2008
R2 Operating System, which hosts the
SANmelody environment and the Hyper-V
basedVDs, and DataCore’s SANmelody itself.
The fully-burdened cost of the benchmarked
configuration, including all of its
infrastructural software and hardware,
but excluding the OS and applications of
theVDs, is thus $67.39 per Desktop.
Scaling – As has been discussed, the
real problem withVDs has been, not in
scaling them up to “thousands” ofVirtual
Desktops but in scaling them down to
practical configurations. It barely needs
mentioning that this must occur without
radically spiking costs at the low end and also
without forgoing the SAN feature set which
assures portability, availability, and data
redundancy. Otherwise, the very benefits
ofVDI are compromised.The DataCore
configuration, described herein, does a good
job of maintaining low infrastructure costs
at 220 Desktops by creating a paired server
approach which simultaneously hosts both
theVDs and the storage infrastructure.
But, what about scaling upward? How does
one arrive at “thousands” of Desktops,
if that is what is required, and more
importantly, is it possible to arrive at
such numbers without wildly varying cost
increments and complex reconfiguration?
To achieve larger configurations, a DataCore
topology known as a “Star” is employed.
With the Star, the addition of a third node
is used to create a Star center or hub.
Additional nodes are then added to the
Star. Each of the nodes of the Star acts as
before, but they mirror their contents to
the hub node rather than to each other.
In turn, the hub becomes the single point
of control for the entire configuration.
With this scheme, each additional server
added as a point to the star brings with it
its own storage devices and infrastructure
as well as itsVDI environment.The
scheme is self tuning until the hub is no
longer able to accommodate secondary
mirror traffic. As an added benefit, most
degenerativeVDI phenomena, such as
“boot storms”, remain contained in quite
small and manageable modules without
dominating the entire topology.
Simulation and initial experimentation
indicates that a simple Star will scale
performance (and perVD costs) to
several thousand Desktops. DataCore
will publish results to bolster these initial
findings when they become available.
5
Conclusion and Future Directions
This paper reports on an architecture for
VDI based on DataCore’s SANmelody,
Microsoft’s Server 2008/Hyper-V,
and standard server hardware. Login
Consultant’sVirtual Session Indexer (VSI)
is used as the benchmarking vehicle.
On the whole, the benchmarked architecture
differs from most of the results published
by other storage vendors. Instead of
constructing aVDI configuration around
a storage controller or pre-existing SAN
architecture, it builds the SAN around the
VDI servers themselves. In particular, the
VDI servers also act as virtualization servers.
The resulting locality produces results with
as much as ten times the price/performance
of other published results, while vastly
reducing the number ofVDs (220) at which
such economics become available.The
whole is accomplished on the simplest,
“lowest common denominator” hardware.
Nevertheless, the environment appears to
be “self-tuning” in that it is resistant to major
reevaluation as it scales upward, while it is
also modular enough to be resilient in the
face of hardware failures, reconfigurations
andVDI anomalies, such as “boot storms”.
Where to from here?
Various optimizations immediately come to
mind. In DataCore’s lab, we have successfully
increased the density ofVDs by more than
50%, to upwards of 175Virtual desktops, on
the same server platforms. Memory tweaks,
enhanced hardware, and various tuning
“tricks” produce the expected results. Still,
increases in costs and complexity produce
a system which is little better in price/
performance than the one tested here.
Sadly, for a group of “old tuners”, we have
had to conclude that the benchmarked
configuration is optimized enough to cross
the threshold for practicalVDI configurations
and that additional low-level optimizations
make only small, incremental changes.
What about high-level optimizations?
Well, perhaps…The co-location of the host
OS, the Hypervisor, and the virtualization
server – alongside theVirtual Desktops – may
provide opportunities for tighter integration.
From our vantage point, however, that these
remain largely independent today offers up
an entirely different class of “optimization”.
VDI has one element in common with the
other major computing movement of our
day: cloud computing. Both technologies
promise to aggregate very large numbers
of “machines” of the same class… that is,
they promise large numbers of platforms,
the resource requirements of which
are similar and which may be predicted
before-hand. Instead of heterogeneous and
unknown requirements, the promise is of
homogeneous and known ones. In our world,
what is “known” is also pre-configurable.
This creates the opportunity to present one
additional level of virtualization and to “divide
and conquer” the one overwhelming fact
which immediately strikes anyone attempting
to work withVDI: the sheer scale of discretely
managing many component machines
which are, nevertheless, irritatingly similar.
What if, instead of attempting to manage
hundreds or thousands of virtual machines
discretely, one could divide them into
arbitrary groups - sub-units, “virtual
datacenters”, or in DataCore parlance,
“hives” – and then treat the entire hive as one
unit? In the this case, a “hive” would consist
of group ofVDs (perhaps 50 or 75 of them),
all grouped around a DataCore virtualization
server and perhaps other servers needed to
provide local services… all implemented as
virtual machines, all configured to interact
with each other and all without regard to any
detailed knowledge about the underlying
hardware environment.The entire hive
would then be administered, moved, and
managed as one unit with the virtualization
server acting as a single “port” to the
outside world. From the standpoint of the
environment, all that would be needed would
be a rough approximation of capacity: two
“hives” over here and four of them, there…
This will be the future direction of our work.
© 2011 DataCore Software Corporation. All Rights Reserved. DataCore, the DataCore logo and SANsymphony are trademarks or registered trademarks of
DataCore Software Corporation. All other products, services and company names mentioned herein may be trademarks of their respective owners.
For additional information,
please visit: www.datacore.com
or e-mail: info@datacore.com
About DataCore Software
DataCore Software develops the storage virtualization software needed to get the highest availability, fastest performance and
maximum utilization from storage resources in physical and virtual IT environments. It provides the critical 3rd dimension on which the
success of server and desktop virtualization projects hinge, regardless of the models and brands of storage devices used.
6
0711
Addendum
BenchmarkTesting and Configuration
The performance benchmark used to
determine the maximum number ofVirtual
Desktops in this configuration wasVirtual
Session Indexer (VSI) rev 2.1.2 by Login
Consultants. LoginVSI creates a simulated
user workload (Outlook, Internet Explorer,
Excel, etc…) on each of the desktops.The
desktops report performance information
to a common share folder which is analyzed
after the completion of the benchmark run.
The simulated user workload is initiated
by opening a Remote Desktop session to
the virtual desktop. In this case, theVSI
Launcher program was used in its default
configuration to open the sessions, one
desktop every 60 seconds.The user workload
runs in a continuous 12 minute loop until all
of the desktop sessions have been opened
and the last one has run through a loop.
After all virtual desktops have run their
simulated user workload simultaneously and
reported their data to a share folder; the
VSI Analysis tool was run.The tool compiles
the data in Excel and produces a graph and
tables showing response times indicating
satisfactory or unsatisfactory user experience.
Configuration Information:
Virtual Desktops -
Microsoft Windows 7 Enterprise x86 –
Configured with the defaultVSI optimizations
and the optimizations recommended in
Virtual Reality Check’s whitepaper “Project
VRC: Phase III” http://www.projectvrc.nl/
Benchmark –
Login ConsultantsVSI 2.1.2
http://www.loginconsultants.com/
PerVSI recommendation, sevenVSI
Launchers were used. All Launchers
were running asVMs on a Hyper-V host
independent from the hosts running
the virtual desktops.The OS of these
VMs was Microsoft Server 2008 R2.
DataCore SANmelody 3.4
http://www.datacore.com/
SANmelody was run with two bug fixes
which will be included in a maintenance
update scheduled for this quarter.
VSI Benchmark Analysis showing a successful run of 220 desktops.

More Related Content

What's hot

Simplified VDI: Dell PowerEdge VRTX & Citrix XenDesktop 7.5
Simplified VDI: Dell PowerEdge VRTX & Citrix XenDesktop 7.5 Simplified VDI: Dell PowerEdge VRTX & Citrix XenDesktop 7.5
Simplified VDI: Dell PowerEdge VRTX & Citrix XenDesktop 7.5 Principled Technologies
 
Comparing performance and cost: Dell PowerEdge VRTX with one Dell PowerEdge M...
Comparing performance and cost: Dell PowerEdge VRTX with one Dell PowerEdge M...Comparing performance and cost: Dell PowerEdge VRTX with one Dell PowerEdge M...
Comparing performance and cost: Dell PowerEdge VRTX with one Dell PowerEdge M...Principled Technologies
 
Vmware documentação tecnica
Vmware documentação tecnicaVmware documentação tecnica
Vmware documentação tecnicaALEXANDRE MEDINA
 
Cost and performance comparison for OpenStack compute and storage infrastructure
Cost and performance comparison for OpenStack compute and storage infrastructureCost and performance comparison for OpenStack compute and storage infrastructure
Cost and performance comparison for OpenStack compute and storage infrastructurePrincipled Technologies
 
Boosting performance with the Dell Acceleration Appliance for Databases
Boosting performance with the Dell Acceleration Appliance for DatabasesBoosting performance with the Dell Acceleration Appliance for Databases
Boosting performance with the Dell Acceleration Appliance for DatabasesPrincipled Technologies
 
Demartek Lenovo Storage S3200 i a mixed workload environment_2016-01
Demartek Lenovo Storage S3200  i a mixed workload environment_2016-01Demartek Lenovo Storage S3200  i a mixed workload environment_2016-01
Demartek Lenovo Storage S3200 i a mixed workload environment_2016-01Lenovo Data Center
 
Intelligent storage management solution using VMware vSphere 5.0 Storage DRS:...
Intelligent storage management solution using VMware vSphere 5.0 Storage DRS:...Intelligent storage management solution using VMware vSphere 5.0 Storage DRS:...
Intelligent storage management solution using VMware vSphere 5.0 Storage DRS:...IBM India Smarter Computing
 
Dynamic Storage Mobility for Hyper V
Dynamic Storage Mobility for Hyper VDynamic Storage Mobility for Hyper V
Dynamic Storage Mobility for Hyper VDataCore Software
 
Configuring a failover cluster on a Dell PowerEdge VRTX
Configuring a failover cluster on a Dell PowerEdge VRTXConfiguring a failover cluster on a Dell PowerEdge VRTX
Configuring a failover cluster on a Dell PowerEdge VRTXPrincipled Technologies
 
TECHNICAL BRIEF▶ Using Virtual Tape Libraries with Backup Exec 2014
TECHNICAL BRIEF▶ Using Virtual Tape Libraries with Backup Exec 2014TECHNICAL BRIEF▶ Using Virtual Tape Libraries with Backup Exec 2014
TECHNICAL BRIEF▶ Using Virtual Tape Libraries with Backup Exec 2014Symantec
 
IBM SONAS and VMware vSphere 5 scale-out cloud foundation: A reference guide ...
IBM SONAS and VMware vSphere 5 scale-out cloud foundation: A reference guide ...IBM SONAS and VMware vSphere 5 scale-out cloud foundation: A reference guide ...
IBM SONAS and VMware vSphere 5 scale-out cloud foundation: A reference guide ...IBM India Smarter Computing
 
Great Article, Thanks Paul Feresten, Sr. Product Marketing Manager, and Rajes...
Great Article, Thanks Paul Feresten, Sr. Product Marketing Manager, and Rajes...Great Article, Thanks Paul Feresten, Sr. Product Marketing Manager, and Rajes...
Great Article, Thanks Paul Feresten, Sr. Product Marketing Manager, and Rajes...Michael Hudak
 
Lenovo Storage S3200 Simple Setup
Lenovo Storage S3200 Simple SetupLenovo Storage S3200 Simple Setup
Lenovo Storage S3200 Simple SetupLenovo Data Center
 
Virtualizing SharePoint Components
Virtualizing SharePoint ComponentsVirtualizing SharePoint Components
Virtualizing SharePoint Componentswebhostingguy
 
Dell PowerEdge VRTX and M-series compute nodes configuration study
Dell PowerEdge VRTX and M-series compute nodes configuration studyDell PowerEdge VRTX and M-series compute nodes configuration study
Dell PowerEdge VRTX and M-series compute nodes configuration studyPrincipled Technologies
 
Storage Considerations for VDI - Scalar presentation at Toronto VMUG 2014
Storage Considerations for VDI - Scalar presentation at Toronto VMUG 2014Storage Considerations for VDI - Scalar presentation at Toronto VMUG 2014
Storage Considerations for VDI - Scalar presentation at Toronto VMUG 2014Scalar Decisions
 
VDI performance comparison: Dell PowerEdge FX2 and FC430 servers with VMware ...
VDI performance comparison: Dell PowerEdge FX2 and FC430 servers with VMware ...VDI performance comparison: Dell PowerEdge FX2 and FC430 servers with VMware ...
VDI performance comparison: Dell PowerEdge FX2 and FC430 servers with VMware ...Principled Technologies
 
on the most suitable storage architecture for virtualization
on the most suitable storage architecture for virtualizationon the most suitable storage architecture for virtualization
on the most suitable storage architecture for virtualizationJordi Moles Blanco
 
Microsoft Windows Server 2012 Early Adopter Guide
Microsoft Windows Server 2012 Early Adopter GuideMicrosoft Windows Server 2012 Early Adopter Guide
Microsoft Windows Server 2012 Early Adopter GuideKingfin Enterprises Limited
 

What's hot (19)

Simplified VDI: Dell PowerEdge VRTX & Citrix XenDesktop 7.5
Simplified VDI: Dell PowerEdge VRTX & Citrix XenDesktop 7.5 Simplified VDI: Dell PowerEdge VRTX & Citrix XenDesktop 7.5
Simplified VDI: Dell PowerEdge VRTX & Citrix XenDesktop 7.5
 
Comparing performance and cost: Dell PowerEdge VRTX with one Dell PowerEdge M...
Comparing performance and cost: Dell PowerEdge VRTX with one Dell PowerEdge M...Comparing performance and cost: Dell PowerEdge VRTX with one Dell PowerEdge M...
Comparing performance and cost: Dell PowerEdge VRTX with one Dell PowerEdge M...
 
Vmware documentação tecnica
Vmware documentação tecnicaVmware documentação tecnica
Vmware documentação tecnica
 
Cost and performance comparison for OpenStack compute and storage infrastructure
Cost and performance comparison for OpenStack compute and storage infrastructureCost and performance comparison for OpenStack compute and storage infrastructure
Cost and performance comparison for OpenStack compute and storage infrastructure
 
Boosting performance with the Dell Acceleration Appliance for Databases
Boosting performance with the Dell Acceleration Appliance for DatabasesBoosting performance with the Dell Acceleration Appliance for Databases
Boosting performance with the Dell Acceleration Appliance for Databases
 
Demartek Lenovo Storage S3200 i a mixed workload environment_2016-01
Demartek Lenovo Storage S3200  i a mixed workload environment_2016-01Demartek Lenovo Storage S3200  i a mixed workload environment_2016-01
Demartek Lenovo Storage S3200 i a mixed workload environment_2016-01
 
Intelligent storage management solution using VMware vSphere 5.0 Storage DRS:...
Intelligent storage management solution using VMware vSphere 5.0 Storage DRS:...Intelligent storage management solution using VMware vSphere 5.0 Storage DRS:...
Intelligent storage management solution using VMware vSphere 5.0 Storage DRS:...
 
Dynamic Storage Mobility for Hyper V
Dynamic Storage Mobility for Hyper VDynamic Storage Mobility for Hyper V
Dynamic Storage Mobility for Hyper V
 
Configuring a failover cluster on a Dell PowerEdge VRTX
Configuring a failover cluster on a Dell PowerEdge VRTXConfiguring a failover cluster on a Dell PowerEdge VRTX
Configuring a failover cluster on a Dell PowerEdge VRTX
 
TECHNICAL BRIEF▶ Using Virtual Tape Libraries with Backup Exec 2014
TECHNICAL BRIEF▶ Using Virtual Tape Libraries with Backup Exec 2014TECHNICAL BRIEF▶ Using Virtual Tape Libraries with Backup Exec 2014
TECHNICAL BRIEF▶ Using Virtual Tape Libraries with Backup Exec 2014
 
IBM SONAS and VMware vSphere 5 scale-out cloud foundation: A reference guide ...
IBM SONAS and VMware vSphere 5 scale-out cloud foundation: A reference guide ...IBM SONAS and VMware vSphere 5 scale-out cloud foundation: A reference guide ...
IBM SONAS and VMware vSphere 5 scale-out cloud foundation: A reference guide ...
 
Great Article, Thanks Paul Feresten, Sr. Product Marketing Manager, and Rajes...
Great Article, Thanks Paul Feresten, Sr. Product Marketing Manager, and Rajes...Great Article, Thanks Paul Feresten, Sr. Product Marketing Manager, and Rajes...
Great Article, Thanks Paul Feresten, Sr. Product Marketing Manager, and Rajes...
 
Lenovo Storage S3200 Simple Setup
Lenovo Storage S3200 Simple SetupLenovo Storage S3200 Simple Setup
Lenovo Storage S3200 Simple Setup
 
Virtualizing SharePoint Components
Virtualizing SharePoint ComponentsVirtualizing SharePoint Components
Virtualizing SharePoint Components
 
Dell PowerEdge VRTX and M-series compute nodes configuration study
Dell PowerEdge VRTX and M-series compute nodes configuration studyDell PowerEdge VRTX and M-series compute nodes configuration study
Dell PowerEdge VRTX and M-series compute nodes configuration study
 
Storage Considerations for VDI - Scalar presentation at Toronto VMUG 2014
Storage Considerations for VDI - Scalar presentation at Toronto VMUG 2014Storage Considerations for VDI - Scalar presentation at Toronto VMUG 2014
Storage Considerations for VDI - Scalar presentation at Toronto VMUG 2014
 
VDI performance comparison: Dell PowerEdge FX2 and FC430 servers with VMware ...
VDI performance comparison: Dell PowerEdge FX2 and FC430 servers with VMware ...VDI performance comparison: Dell PowerEdge FX2 and FC430 servers with VMware ...
VDI performance comparison: Dell PowerEdge FX2 and FC430 servers with VMware ...
 
on the most suitable storage architecture for virtualization
on the most suitable storage architecture for virtualizationon the most suitable storage architecture for virtualization
on the most suitable storage architecture for virtualization
 
Microsoft Windows Server 2012 Early Adopter Guide
Microsoft Windows Server 2012 Early Adopter GuideMicrosoft Windows Server 2012 Early Adopter Guide
Microsoft Windows Server 2012 Early Adopter Guide
 

Viewers also liked

Virtualized SANS for Virtual Servers
Virtualized SANS for Virtual ServersVirtualized SANS for Virtual Servers
Virtualized SANS for Virtual ServersDataCore Software
 
Software Defined Storage In Action
Software Defined Storage In ActionSoftware Defined Storage In Action
Software Defined Storage In ActionDataCore Software
 
Addressing the Top 3 Storage Challenges in Healthcare with Hanover Hospital
Addressing the Top 3 Storage Challenges in Healthcare with Hanover HospitalAddressing the Top 3 Storage Challenges in Healthcare with Hanover Hospital
Addressing the Top 3 Storage Challenges in Healthcare with Hanover HospitalDataCore Software
 
ESG Datacore SANsymphony-V Whitepaper
ESG Datacore SANsymphony-V WhitepaperESG Datacore SANsymphony-V Whitepaper
ESG Datacore SANsymphony-V WhitepaperDataCore Software
 
Insider's Guide- Building a Virtualized Storage Service
Insider's Guide- Building a Virtualized Storage ServiceInsider's Guide- Building a Virtualized Storage Service
Insider's Guide- Building a Virtualized Storage ServiceDataCore Software
 
Delivering First Class performance and Availability for Virtualized Tier 1 Apps
Delivering First Class performance and Availability for Virtualized Tier 1 Apps Delivering First Class performance and Availability for Virtualized Tier 1 Apps
Delivering First Class performance and Availability for Virtualized Tier 1 Apps DataCore Software
 
Optimizing The Economics of Storage: It's All About the Benjamins
Optimizing The Economics of Storage: It's All About the BenjaminsOptimizing The Economics of Storage: It's All About the Benjamins
Optimizing The Economics of Storage: It's All About the BenjaminsDataCore Software
 
Increase Your Mission Critical Application Performance without Breaking the B...
Increase Your Mission Critical Application Performance without Breaking the B...Increase Your Mission Critical Application Performance without Breaking the B...
Increase Your Mission Critical Application Performance without Breaking the B...DataCore Software
 
Uninterrupted access to Cluster Shared volumes (CSVs) Synchronously Mirrored ...
Uninterrupted access to Cluster Shared volumes (CSVs) Synchronously Mirrored ...Uninterrupted access to Cluster Shared volumes (CSVs) Synchronously Mirrored ...
Uninterrupted access to Cluster Shared volumes (CSVs) Synchronously Mirrored ...DataCore Software
 
Can $0.08 Change your View of Storage?
Can $0.08 Change your View of Storage?Can $0.08 Change your View of Storage?
Can $0.08 Change your View of Storage?DataCore Software
 
App Performance Tip: Sharing Flash Across Virtualized Workloads
App Performance Tip: Sharing Flash Across Virtualized WorkloadsApp Performance Tip: Sharing Flash Across Virtualized Workloads
App Performance Tip: Sharing Flash Across Virtualized WorkloadsDataCore Software
 
The Storage Side of Private Clouds
The Storage Side of Private CloudsThe Storage Side of Private Clouds
The Storage Side of Private CloudsDataCore Software
 
Removing Storage Related Barriers to Server and Desktop Virtualization
Removing Storage Related Barriers to Server and Desktop VirtualizationRemoving Storage Related Barriers to Server and Desktop Virtualization
Removing Storage Related Barriers to Server and Desktop VirtualizationDataCore Software
 
How to Sell Storage Virtualization to The CIO
How to Sell Storage Virtualization to The CIOHow to Sell Storage Virtualization to The CIO
How to Sell Storage Virtualization to The CIODataCore Software
 
Software-defined Storage Makes Economic Sense
Software-defined Storage Makes Economic SenseSoftware-defined Storage Makes Economic Sense
Software-defined Storage Makes Economic SenseDataCore Software
 
Virtual SAN vs Good Old SANs: Can't they just get along?
Virtual SAN vs Good Old SANs: Can't they just get along?Virtual SAN vs Good Old SANs: Can't they just get along?
Virtual SAN vs Good Old SANs: Can't they just get along?DataCore Software
 
Virtual SAN: It’s a SAN, it’s Virtual, but what is it really?
Virtual SAN: It’s a SAN, it’s Virtual, but what is it really?Virtual SAN: It’s a SAN, it’s Virtual, but what is it really?
Virtual SAN: It’s a SAN, it’s Virtual, but what is it really?DataCore Software
 
Creating An Enterprise Storage Strategy for Healthcare
Creating An Enterprise Storage Strategy for Healthcare  Creating An Enterprise Storage Strategy for Healthcare
Creating An Enterprise Storage Strategy for Healthcare DataCore Software
 

Viewers also liked (20)

Virtualized SANS for Virtual Servers
Virtualized SANS for Virtual ServersVirtualized SANS for Virtual Servers
Virtualized SANS for Virtual Servers
 
Software Defined Storage In Action
Software Defined Storage In ActionSoftware Defined Storage In Action
Software Defined Storage In Action
 
Addressing the Top 3 Storage Challenges in Healthcare with Hanover Hospital
Addressing the Top 3 Storage Challenges in Healthcare with Hanover HospitalAddressing the Top 3 Storage Challenges in Healthcare with Hanover Hospital
Addressing the Top 3 Storage Challenges in Healthcare with Hanover Hospital
 
ESG Datacore SANsymphony-V Whitepaper
ESG Datacore SANsymphony-V WhitepaperESG Datacore SANsymphony-V Whitepaper
ESG Datacore SANsymphony-V Whitepaper
 
Insider's Guide- Building a Virtualized Storage Service
Insider's Guide- Building a Virtualized Storage ServiceInsider's Guide- Building a Virtualized Storage Service
Insider's Guide- Building a Virtualized Storage Service
 
Delivering First Class performance and Availability for Virtualized Tier 1 Apps
Delivering First Class performance and Availability for Virtualized Tier 1 Apps Delivering First Class performance and Availability for Virtualized Tier 1 Apps
Delivering First Class performance and Availability for Virtualized Tier 1 Apps
 
Solutions for Healthcare IT
Solutions for Healthcare ITSolutions for Healthcare IT
Solutions for Healthcare IT
 
Optimizing The Economics of Storage: It's All About the Benjamins
Optimizing The Economics of Storage: It's All About the BenjaminsOptimizing The Economics of Storage: It's All About the Benjamins
Optimizing The Economics of Storage: It's All About the Benjamins
 
Increase Your Mission Critical Application Performance without Breaking the B...
Increase Your Mission Critical Application Performance without Breaking the B...Increase Your Mission Critical Application Performance without Breaking the B...
Increase Your Mission Critical Application Performance without Breaking the B...
 
Uninterrupted access to Cluster Shared volumes (CSVs) Synchronously Mirrored ...
Uninterrupted access to Cluster Shared volumes (CSVs) Synchronously Mirrored ...Uninterrupted access to Cluster Shared volumes (CSVs) Synchronously Mirrored ...
Uninterrupted access to Cluster Shared volumes (CSVs) Synchronously Mirrored ...
 
Can $0.08 Change your View of Storage?
Can $0.08 Change your View of Storage?Can $0.08 Change your View of Storage?
Can $0.08 Change your View of Storage?
 
App Performance Tip: Sharing Flash Across Virtualized Workloads
App Performance Tip: Sharing Flash Across Virtualized WorkloadsApp Performance Tip: Sharing Flash Across Virtualized Workloads
App Performance Tip: Sharing Flash Across Virtualized Workloads
 
Automated Storage Tiering
Automated Storage TieringAutomated Storage Tiering
Automated Storage Tiering
 
The Storage Side of Private Clouds
The Storage Side of Private CloudsThe Storage Side of Private Clouds
The Storage Side of Private Clouds
 
Removing Storage Related Barriers to Server and Desktop Virtualization
Removing Storage Related Barriers to Server and Desktop VirtualizationRemoving Storage Related Barriers to Server and Desktop Virtualization
Removing Storage Related Barriers to Server and Desktop Virtualization
 
How to Sell Storage Virtualization to The CIO
How to Sell Storage Virtualization to The CIOHow to Sell Storage Virtualization to The CIO
How to Sell Storage Virtualization to The CIO
 
Software-defined Storage Makes Economic Sense
Software-defined Storage Makes Economic SenseSoftware-defined Storage Makes Economic Sense
Software-defined Storage Makes Economic Sense
 
Virtual SAN vs Good Old SANs: Can't they just get along?
Virtual SAN vs Good Old SANs: Can't they just get along?Virtual SAN vs Good Old SANs: Can't they just get along?
Virtual SAN vs Good Old SANs: Can't they just get along?
 
Virtual SAN: It’s a SAN, it’s Virtual, but what is it really?
Virtual SAN: It’s a SAN, it’s Virtual, but what is it really?Virtual SAN: It’s a SAN, it’s Virtual, but what is it really?
Virtual SAN: It’s a SAN, it’s Virtual, but what is it really?
 
Creating An Enterprise Storage Strategy for Healthcare
Creating An Enterprise Storage Strategy for Healthcare  Creating An Enterprise Storage Strategy for Healthcare
Creating An Enterprise Storage Strategy for Healthcare
 

Similar to Benchmarking a Scalable and Highly Available Architecture for Virtual Desktops

Business-critical applications on VMware vSphere 6, VMware Virtual SAN, and V...
Business-critical applications on VMware vSphere 6, VMware Virtual SAN, and V...Business-critical applications on VMware vSphere 6, VMware Virtual SAN, and V...
Business-critical applications on VMware vSphere 6, VMware Virtual SAN, and V...Principled Technologies
 
Complemenatry & Alternative Solution to VDI
Complemenatry & Alternative Solution to VDIComplemenatry & Alternative Solution to VDI
Complemenatry & Alternative Solution to VDIDeskStream, Inc
 
Gridstore's Software-Defined-Storage Architecture
Gridstore's Software-Defined-Storage ArchitectureGridstore's Software-Defined-Storage Architecture
Gridstore's Software-Defined-Storage ArchitectureGridstore
 
Build your private cloud with paa s using linuxz cover story enterprise tech ...
Build your private cloud with paa s using linuxz cover story enterprise tech ...Build your private cloud with paa s using linuxz cover story enterprise tech ...
Build your private cloud with paa s using linuxz cover story enterprise tech ...Elena Nanos
 
Renaissance in VM Network Connectivity
Renaissance in VM Network ConnectivityRenaissance in VM Network Connectivity
Renaissance in VM Network ConnectivityIT Brand Pulse
 
VDI with Dell EqualLogic hybrid arrays: A comparative study with the industry...
VDI with Dell EqualLogic hybrid arrays: A comparative study with the industry...VDI with Dell EqualLogic hybrid arrays: A comparative study with the industry...
VDI with Dell EqualLogic hybrid arrays: A comparative study with the industry...Principled Technologies
 
Hyper-converged infrastructure
Hyper-converged infrastructureHyper-converged infrastructure
Hyper-converged infrastructureIgor Malts
 
The Best Infrastructure for OpenStack: VMware vSphere and Virtual SAN
The Best Infrastructure for OpenStack: VMware vSphere and Virtual SANThe Best Infrastructure for OpenStack: VMware vSphere and Virtual SAN
The Best Infrastructure for OpenStack: VMware vSphere and Virtual SANEMC
 
VMmark 2.5.2 virtualization performance of the Dell EqualLogic PS6210XS stora...
VMmark 2.5.2 virtualization performance of the Dell EqualLogic PS6210XS stora...VMmark 2.5.2 virtualization performance of the Dell EqualLogic PS6210XS stora...
VMmark 2.5.2 virtualization performance of the Dell EqualLogic PS6210XS stora...Principled Technologies
 
Virtulization
VirtulizationVirtulization
VirtulizationAntu Sabu
 
How to build a winning solution for large scale VDI deployments
How to build a winning solution for large scale VDI deploymentsHow to build a winning solution for large scale VDI deployments
How to build a winning solution for large scale VDI deploymentsNetApp
 
A Survey of Performance Comparison between Virtual Machines and Containers
A Survey of Performance Comparison between Virtual Machines and ContainersA Survey of Performance Comparison between Virtual Machines and Containers
A Survey of Performance Comparison between Virtual Machines and Containersprashant desai
 
Datacenter migration using vmware
Datacenter migration using vmwareDatacenter migration using vmware
Datacenter migration using vmwareWilson Erique
 
Exploring HCI
Exploring HCIExploring HCI
Exploring HCIDavid Han
 
Storage Strategies Now- Virtualizaing Busines Critical applications
Storage Strategies Now- Virtualizaing Busines Critical applicationsStorage Strategies Now- Virtualizaing Busines Critical applications
Storage Strategies Now- Virtualizaing Busines Critical applicationsDataCore Software
 
LIQUID-A Scalable Deduplication File System For Virtual Machine Images
LIQUID-A Scalable Deduplication File System For Virtual Machine ImagesLIQUID-A Scalable Deduplication File System For Virtual Machine Images
LIQUID-A Scalable Deduplication File System For Virtual Machine Imagesfabna benz
 
Boosting virtualization performance with Intel SSD DC Series P3600 NVMe SSDs ...
Boosting virtualization performance with Intel SSD DC Series P3600 NVMe SSDs ...Boosting virtualization performance with Intel SSD DC Series P3600 NVMe SSDs ...
Boosting virtualization performance with Intel SSD DC Series P3600 NVMe SSDs ...Principled Technologies
 
Hardware VDI vs. Software VDI
Hardware VDI vs. Software VDIHardware VDI vs. Software VDI
Hardware VDI vs. Software VDIcitrixgurl
 

Similar to Benchmarking a Scalable and Highly Available Architecture for Virtual Desktops (20)

Business-critical applications on VMware vSphere 6, VMware Virtual SAN, and V...
Business-critical applications on VMware vSphere 6, VMware Virtual SAN, and V...Business-critical applications on VMware vSphere 6, VMware Virtual SAN, and V...
Business-critical applications on VMware vSphere 6, VMware Virtual SAN, and V...
 
Complemenatry & Alternative Solution to VDI
Complemenatry & Alternative Solution to VDIComplemenatry & Alternative Solution to VDI
Complemenatry & Alternative Solution to VDI
 
Gridstore's Software-Defined-Storage Architecture
Gridstore's Software-Defined-Storage ArchitectureGridstore's Software-Defined-Storage Architecture
Gridstore's Software-Defined-Storage Architecture
 
Cloud Technology: Virtualization
Cloud Technology: VirtualizationCloud Technology: Virtualization
Cloud Technology: Virtualization
 
Build your private cloud with paa s using linuxz cover story enterprise tech ...
Build your private cloud with paa s using linuxz cover story enterprise tech ...Build your private cloud with paa s using linuxz cover story enterprise tech ...
Build your private cloud with paa s using linuxz cover story enterprise tech ...
 
Renaissance in VM Network Connectivity
Renaissance in VM Network ConnectivityRenaissance in VM Network Connectivity
Renaissance in VM Network Connectivity
 
VDI with Dell EqualLogic hybrid arrays: A comparative study with the industry...
VDI with Dell EqualLogic hybrid arrays: A comparative study with the industry...VDI with Dell EqualLogic hybrid arrays: A comparative study with the industry...
VDI with Dell EqualLogic hybrid arrays: A comparative study with the industry...
 
Hyper-converged infrastructure
Hyper-converged infrastructureHyper-converged infrastructure
Hyper-converged infrastructure
 
The Best Infrastructure for OpenStack: VMware vSphere and Virtual SAN
The Best Infrastructure for OpenStack: VMware vSphere and Virtual SANThe Best Infrastructure for OpenStack: VMware vSphere and Virtual SAN
The Best Infrastructure for OpenStack: VMware vSphere and Virtual SAN
 
VMmark 2.5.2 virtualization performance of the Dell EqualLogic PS6210XS stora...
VMmark 2.5.2 virtualization performance of the Dell EqualLogic PS6210XS stora...VMmark 2.5.2 virtualization performance of the Dell EqualLogic PS6210XS stora...
VMmark 2.5.2 virtualization performance of the Dell EqualLogic PS6210XS stora...
 
Virtulization
VirtulizationVirtulization
Virtulization
 
How to build a winning solution for large scale VDI deployments
How to build a winning solution for large scale VDI deploymentsHow to build a winning solution for large scale VDI deployments
How to build a winning solution for large scale VDI deployments
 
VMware 2009
VMware 2009VMware 2009
VMware 2009
 
A Survey of Performance Comparison between Virtual Machines and Containers
A Survey of Performance Comparison between Virtual Machines and ContainersA Survey of Performance Comparison between Virtual Machines and Containers
A Survey of Performance Comparison between Virtual Machines and Containers
 
Datacenter migration using vmware
Datacenter migration using vmwareDatacenter migration using vmware
Datacenter migration using vmware
 
Exploring HCI
Exploring HCIExploring HCI
Exploring HCI
 
Storage Strategies Now- Virtualizaing Busines Critical applications
Storage Strategies Now- Virtualizaing Busines Critical applicationsStorage Strategies Now- Virtualizaing Busines Critical applications
Storage Strategies Now- Virtualizaing Busines Critical applications
 
LIQUID-A Scalable Deduplication File System For Virtual Machine Images
LIQUID-A Scalable Deduplication File System For Virtual Machine ImagesLIQUID-A Scalable Deduplication File System For Virtual Machine Images
LIQUID-A Scalable Deduplication File System For Virtual Machine Images
 
Boosting virtualization performance with Intel SSD DC Series P3600 NVMe SSDs ...
Boosting virtualization performance with Intel SSD DC Series P3600 NVMe SSDs ...Boosting virtualization performance with Intel SSD DC Series P3600 NVMe SSDs ...
Boosting virtualization performance with Intel SSD DC Series P3600 NVMe SSDs ...
 
Hardware VDI vs. Software VDI
Hardware VDI vs. Software VDIHardware VDI vs. Software VDI
Hardware VDI vs. Software VDI
 

More from DataCore Software

Software-Defined Storage Accelerates Storage Cost Reduction and Service-Level...
Software-Defined Storage Accelerates Storage Cost Reduction and Service-Level...Software-Defined Storage Accelerates Storage Cost Reduction and Service-Level...
Software-Defined Storage Accelerates Storage Cost Reduction and Service-Level...DataCore Software
 
NVMe and Flash – Make Your Storage Great Again!
NVMe and Flash – Make Your Storage Great Again!NVMe and Flash – Make Your Storage Great Again!
NVMe and Flash – Make Your Storage Great Again!DataCore Software
 
Zero Downtime, Zero Touch Stretch Clusters from Software-Defined Storage
Zero Downtime, Zero Touch Stretch Clusters from Software-Defined StorageZero Downtime, Zero Touch Stretch Clusters from Software-Defined Storage
Zero Downtime, Zero Touch Stretch Clusters from Software-Defined StorageDataCore Software
 
From Disaster to Recovery: Preparing Your IT for the Unexpected
From Disaster to Recovery: Preparing Your IT for the UnexpectedFrom Disaster to Recovery: Preparing Your IT for the Unexpected
From Disaster to Recovery: Preparing Your IT for the UnexpectedDataCore Software
 
How to Integrate Hyperconverged Systems with Existing SANs
How to Integrate Hyperconverged Systems with Existing SANsHow to Integrate Hyperconverged Systems with Existing SANs
How to Integrate Hyperconverged Systems with Existing SANsDataCore Software
 
How to Avoid Disasters via Software-Defined Storage Replication & Site Recovery
How to Avoid Disasters via Software-Defined Storage Replication & Site RecoveryHow to Avoid Disasters via Software-Defined Storage Replication & Site Recovery
How to Avoid Disasters via Software-Defined Storage Replication & Site RecoveryDataCore Software
 
Cloud Infrastructure for Your Data Center
Cloud Infrastructure for Your Data CenterCloud Infrastructure for Your Data Center
Cloud Infrastructure for Your Data CenterDataCore Software
 
Building a Highly Available Data Infrastructure
Building a Highly Available Data InfrastructureBuilding a Highly Available Data Infrastructure
Building a Highly Available Data InfrastructureDataCore Software
 
Top 3 Challenges Impacting Your Data and How to Solve Them
Top 3 Challenges Impacting Your Data and How to Solve ThemTop 3 Challenges Impacting Your Data and How to Solve Them
Top 3 Challenges Impacting Your Data and How to Solve ThemDataCore Software
 
Business Continuity for Mission Critical Applications
Business Continuity for Mission Critical ApplicationsBusiness Continuity for Mission Critical Applications
Business Continuity for Mission Critical ApplicationsDataCore Software
 
Dynamic Hyper-Converged Future Proof Your Data Center
Dynamic Hyper-Converged Future Proof Your Data CenterDynamic Hyper-Converged Future Proof Your Data Center
Dynamic Hyper-Converged Future Proof Your Data CenterDataCore Software
 
Community Health Network Delivers Unprecedented Availability for Critical Hea...
Community Health Network Delivers Unprecedented Availability for Critical Hea...Community Health Network Delivers Unprecedented Availability for Critical Hea...
Community Health Network Delivers Unprecedented Availability for Critical Hea...DataCore Software
 
Case Study: Mission Community Hospital
Case Study: Mission Community HospitalCase Study: Mission Community Hospital
Case Study: Mission Community HospitalDataCore Software
 
Emergency Communication of Southern Oregon
Emergency Communication of Southern OregonEmergency Communication of Southern Oregon
Emergency Communication of Southern OregonDataCore Software
 
Integrating Hyper-converged Systems with Existing SANs
Integrating Hyper-converged Systems with Existing SANs Integrating Hyper-converged Systems with Existing SANs
Integrating Hyper-converged Systems with Existing SANs DataCore Software
 
Fighting the Hidden Costs of Data Storage
Fighting the Hidden Costs of Data StorageFighting the Hidden Costs of Data Storage
Fighting the Hidden Costs of Data StorageDataCore Software
 
The Need for Speed: Parallel I/O and the New Tick-Tock in Computing
The Need for Speed: Parallel I/O and the New Tick-Tock in ComputingThe Need for Speed: Parallel I/O and the New Tick-Tock in Computing
The Need for Speed: Parallel I/O and the New Tick-Tock in ComputingDataCore Software
 

More from DataCore Software (20)

Software-Defined Storage Accelerates Storage Cost Reduction and Service-Level...
Software-Defined Storage Accelerates Storage Cost Reduction and Service-Level...Software-Defined Storage Accelerates Storage Cost Reduction and Service-Level...
Software-Defined Storage Accelerates Storage Cost Reduction and Service-Level...
 
NVMe and Flash – Make Your Storage Great Again!
NVMe and Flash – Make Your Storage Great Again!NVMe and Flash – Make Your Storage Great Again!
NVMe and Flash – Make Your Storage Great Again!
 
Zero Downtime, Zero Touch Stretch Clusters from Software-Defined Storage
Zero Downtime, Zero Touch Stretch Clusters from Software-Defined StorageZero Downtime, Zero Touch Stretch Clusters from Software-Defined Storage
Zero Downtime, Zero Touch Stretch Clusters from Software-Defined Storage
 
From Disaster to Recovery: Preparing Your IT for the Unexpected
From Disaster to Recovery: Preparing Your IT for the UnexpectedFrom Disaster to Recovery: Preparing Your IT for the Unexpected
From Disaster to Recovery: Preparing Your IT for the Unexpected
 
How to Integrate Hyperconverged Systems with Existing SANs
How to Integrate Hyperconverged Systems with Existing SANsHow to Integrate Hyperconverged Systems with Existing SANs
How to Integrate Hyperconverged Systems with Existing SANs
 
How to Avoid Disasters via Software-Defined Storage Replication & Site Recovery
How to Avoid Disasters via Software-Defined Storage Replication & Site RecoveryHow to Avoid Disasters via Software-Defined Storage Replication & Site Recovery
How to Avoid Disasters via Software-Defined Storage Replication & Site Recovery
 
Cloud Infrastructure for Your Data Center
Cloud Infrastructure for Your Data CenterCloud Infrastructure for Your Data Center
Cloud Infrastructure for Your Data Center
 
Building a Highly Available Data Infrastructure
Building a Highly Available Data InfrastructureBuilding a Highly Available Data Infrastructure
Building a Highly Available Data Infrastructure
 
TUI Case Study
TUI Case StudyTUI Case Study
TUI Case Study
 
Thorntons Case Study
Thorntons Case StudyThorntons Case Study
Thorntons Case Study
 
Top 3 Challenges Impacting Your Data and How to Solve Them
Top 3 Challenges Impacting Your Data and How to Solve ThemTop 3 Challenges Impacting Your Data and How to Solve Them
Top 3 Challenges Impacting Your Data and How to Solve Them
 
Business Continuity for Mission Critical Applications
Business Continuity for Mission Critical ApplicationsBusiness Continuity for Mission Critical Applications
Business Continuity for Mission Critical Applications
 
Dynamic Hyper-Converged Future Proof Your Data Center
Dynamic Hyper-Converged Future Proof Your Data CenterDynamic Hyper-Converged Future Proof Your Data Center
Dynamic Hyper-Converged Future Proof Your Data Center
 
Community Health Network Delivers Unprecedented Availability for Critical Hea...
Community Health Network Delivers Unprecedented Availability for Critical Hea...Community Health Network Delivers Unprecedented Availability for Critical Hea...
Community Health Network Delivers Unprecedented Availability for Critical Hea...
 
Case Study: Mission Community Hospital
Case Study: Mission Community HospitalCase Study: Mission Community Hospital
Case Study: Mission Community Hospital
 
Emergency Communication of Southern Oregon
Emergency Communication of Southern OregonEmergency Communication of Southern Oregon
Emergency Communication of Southern Oregon
 
DataCore At VMworld 2016
DataCore At VMworld 2016DataCore At VMworld 2016
DataCore At VMworld 2016
 
Integrating Hyper-converged Systems with Existing SANs
Integrating Hyper-converged Systems with Existing SANs Integrating Hyper-converged Systems with Existing SANs
Integrating Hyper-converged Systems with Existing SANs
 
Fighting the Hidden Costs of Data Storage
Fighting the Hidden Costs of Data StorageFighting the Hidden Costs of Data Storage
Fighting the Hidden Costs of Data Storage
 
The Need for Speed: Parallel I/O and the New Tick-Tock in Computing
The Need for Speed: Parallel I/O and the New Tick-Tock in ComputingThe Need for Speed: Parallel I/O and the New Tick-Tock in Computing
The Need for Speed: Parallel I/O and the New Tick-Tock in Computing
 

Recently uploaded

How we scaled to 80K users by doing nothing!.pdf
How we scaled to 80K users by doing nothing!.pdfHow we scaled to 80K users by doing nothing!.pdf
How we scaled to 80K users by doing nothing!.pdfSrushith Repakula
 
Optimizing NoSQL Performance Through Observability
Optimizing NoSQL Performance Through ObservabilityOptimizing NoSQL Performance Through Observability
Optimizing NoSQL Performance Through ObservabilityScyllaDB
 
Speed Wins: From Kafka to APIs in Minutes
Speed Wins: From Kafka to APIs in MinutesSpeed Wins: From Kafka to APIs in Minutes
Speed Wins: From Kafka to APIs in Minutesconfluent
 
Powerful Start- the Key to Project Success, Barbara Laskowska
Powerful Start- the Key to Project Success, Barbara LaskowskaPowerful Start- the Key to Project Success, Barbara Laskowska
Powerful Start- the Key to Project Success, Barbara LaskowskaCzechDreamin
 
WebAssembly is Key to Better LLM Performance
WebAssembly is Key to Better LLM PerformanceWebAssembly is Key to Better LLM Performance
WebAssembly is Key to Better LLM PerformanceSamy Fodil
 
PLAI - Acceleration Program for Generative A.I. Startups
PLAI - Acceleration Program for Generative A.I. StartupsPLAI - Acceleration Program for Generative A.I. Startups
PLAI - Acceleration Program for Generative A.I. StartupsStefano
 
A Business-Centric Approach to Design System Strategy
A Business-Centric Approach to Design System StrategyA Business-Centric Approach to Design System Strategy
A Business-Centric Approach to Design System StrategyUXDXConf
 
FDO for Camera, Sensor and Networking Device – Commercial Solutions from VinC...
FDO for Camera, Sensor and Networking Device – Commercial Solutions from VinC...FDO for Camera, Sensor and Networking Device – Commercial Solutions from VinC...
FDO for Camera, Sensor and Networking Device – Commercial Solutions from VinC...FIDO Alliance
 
AI presentation and introduction - Retrieval Augmented Generation RAG 101
AI presentation and introduction - Retrieval Augmented Generation RAG 101AI presentation and introduction - Retrieval Augmented Generation RAG 101
AI presentation and introduction - Retrieval Augmented Generation RAG 101vincent683379
 
Extensible Python: Robustness through Addition - PyCon 2024
Extensible Python: Robustness through Addition - PyCon 2024Extensible Python: Robustness through Addition - PyCon 2024
Extensible Python: Robustness through Addition - PyCon 2024Patrick Viafore
 
Intro in Product Management - Коротко про професію продакт менеджера
Intro in Product Management - Коротко про професію продакт менеджераIntro in Product Management - Коротко про професію продакт менеджера
Intro in Product Management - Коротко про професію продакт менеджераMark Opanasiuk
 
Linux Foundation Edge _ Overview of FDO Software Components _ Randy at Intel.pdf
Linux Foundation Edge _ Overview of FDO Software Components _ Randy at Intel.pdfLinux Foundation Edge _ Overview of FDO Software Components _ Randy at Intel.pdf
Linux Foundation Edge _ Overview of FDO Software Components _ Randy at Intel.pdfFIDO Alliance
 
Breaking Down the Flutterwave Scandal What You Need to Know.pdf
Breaking Down the Flutterwave Scandal What You Need to Know.pdfBreaking Down the Flutterwave Scandal What You Need to Know.pdf
Breaking Down the Flutterwave Scandal What You Need to Know.pdfUK Journal
 
Secure Zero Touch enabled Edge compute with Dell NativeEdge via FDO _ Brad at...
Secure Zero Touch enabled Edge compute with Dell NativeEdge via FDO _ Brad at...Secure Zero Touch enabled Edge compute with Dell NativeEdge via FDO _ Brad at...
Secure Zero Touch enabled Edge compute with Dell NativeEdge via FDO _ Brad at...FIDO Alliance
 
IESVE for Early Stage Design and Planning
IESVE for Early Stage Design and PlanningIESVE for Early Stage Design and Planning
IESVE for Early Stage Design and PlanningIES VE
 
Microsoft CSP Briefing Pre-Engagement - Questionnaire
Microsoft CSP Briefing Pre-Engagement - QuestionnaireMicrosoft CSP Briefing Pre-Engagement - Questionnaire
Microsoft CSP Briefing Pre-Engagement - QuestionnaireExakis Nelite
 
The Value of Certifying Products for FDO _ Paul at FIDO Alliance.pdf
The Value of Certifying Products for FDO _ Paul at FIDO Alliance.pdfThe Value of Certifying Products for FDO _ Paul at FIDO Alliance.pdf
The Value of Certifying Products for FDO _ Paul at FIDO Alliance.pdfFIDO Alliance
 
Using IESVE for Room Loads Analysis - UK & Ireland
Using IESVE for Room Loads Analysis - UK & IrelandUsing IESVE for Room Loads Analysis - UK & Ireland
Using IESVE for Room Loads Analysis - UK & IrelandIES VE
 
1111 ChatGPT Prompts PDF Free Download - Prompts for ChatGPT
1111 ChatGPT Prompts PDF Free Download - Prompts for ChatGPT1111 ChatGPT Prompts PDF Free Download - Prompts for ChatGPT
1111 ChatGPT Prompts PDF Free Download - Prompts for ChatGPTiSEO AI
 
Structuring Teams and Portfolios for Success
Structuring Teams and Portfolios for SuccessStructuring Teams and Portfolios for Success
Structuring Teams and Portfolios for SuccessUXDXConf
 

Recently uploaded (20)

How we scaled to 80K users by doing nothing!.pdf
How we scaled to 80K users by doing nothing!.pdfHow we scaled to 80K users by doing nothing!.pdf
How we scaled to 80K users by doing nothing!.pdf
 
Optimizing NoSQL Performance Through Observability
Optimizing NoSQL Performance Through ObservabilityOptimizing NoSQL Performance Through Observability
Optimizing NoSQL Performance Through Observability
 
Speed Wins: From Kafka to APIs in Minutes
Speed Wins: From Kafka to APIs in MinutesSpeed Wins: From Kafka to APIs in Minutes
Speed Wins: From Kafka to APIs in Minutes
 
Powerful Start- the Key to Project Success, Barbara Laskowska
Powerful Start- the Key to Project Success, Barbara LaskowskaPowerful Start- the Key to Project Success, Barbara Laskowska
Powerful Start- the Key to Project Success, Barbara Laskowska
 
WebAssembly is Key to Better LLM Performance
WebAssembly is Key to Better LLM PerformanceWebAssembly is Key to Better LLM Performance
WebAssembly is Key to Better LLM Performance
 
PLAI - Acceleration Program for Generative A.I. Startups
PLAI - Acceleration Program for Generative A.I. StartupsPLAI - Acceleration Program for Generative A.I. Startups
PLAI - Acceleration Program for Generative A.I. Startups
 
A Business-Centric Approach to Design System Strategy
A Business-Centric Approach to Design System StrategyA Business-Centric Approach to Design System Strategy
A Business-Centric Approach to Design System Strategy
 
FDO for Camera, Sensor and Networking Device – Commercial Solutions from VinC...
FDO for Camera, Sensor and Networking Device – Commercial Solutions from VinC...FDO for Camera, Sensor and Networking Device – Commercial Solutions from VinC...
FDO for Camera, Sensor and Networking Device – Commercial Solutions from VinC...
 
AI presentation and introduction - Retrieval Augmented Generation RAG 101
AI presentation and introduction - Retrieval Augmented Generation RAG 101AI presentation and introduction - Retrieval Augmented Generation RAG 101
AI presentation and introduction - Retrieval Augmented Generation RAG 101
 
Extensible Python: Robustness through Addition - PyCon 2024
Extensible Python: Robustness through Addition - PyCon 2024Extensible Python: Robustness through Addition - PyCon 2024
Extensible Python: Robustness through Addition - PyCon 2024
 
Intro in Product Management - Коротко про професію продакт менеджера
Intro in Product Management - Коротко про професію продакт менеджераIntro in Product Management - Коротко про професію продакт менеджера
Intro in Product Management - Коротко про професію продакт менеджера
 
Linux Foundation Edge _ Overview of FDO Software Components _ Randy at Intel.pdf
Linux Foundation Edge _ Overview of FDO Software Components _ Randy at Intel.pdfLinux Foundation Edge _ Overview of FDO Software Components _ Randy at Intel.pdf
Linux Foundation Edge _ Overview of FDO Software Components _ Randy at Intel.pdf
 
Breaking Down the Flutterwave Scandal What You Need to Know.pdf
Breaking Down the Flutterwave Scandal What You Need to Know.pdfBreaking Down the Flutterwave Scandal What You Need to Know.pdf
Breaking Down the Flutterwave Scandal What You Need to Know.pdf
 
Secure Zero Touch enabled Edge compute with Dell NativeEdge via FDO _ Brad at...
Secure Zero Touch enabled Edge compute with Dell NativeEdge via FDO _ Brad at...Secure Zero Touch enabled Edge compute with Dell NativeEdge via FDO _ Brad at...
Secure Zero Touch enabled Edge compute with Dell NativeEdge via FDO _ Brad at...
 
IESVE for Early Stage Design and Planning
IESVE for Early Stage Design and PlanningIESVE for Early Stage Design and Planning
IESVE for Early Stage Design and Planning
 
Microsoft CSP Briefing Pre-Engagement - Questionnaire
Microsoft CSP Briefing Pre-Engagement - QuestionnaireMicrosoft CSP Briefing Pre-Engagement - Questionnaire
Microsoft CSP Briefing Pre-Engagement - Questionnaire
 
The Value of Certifying Products for FDO _ Paul at FIDO Alliance.pdf
The Value of Certifying Products for FDO _ Paul at FIDO Alliance.pdfThe Value of Certifying Products for FDO _ Paul at FIDO Alliance.pdf
The Value of Certifying Products for FDO _ Paul at FIDO Alliance.pdf
 
Using IESVE for Room Loads Analysis - UK & Ireland
Using IESVE for Room Loads Analysis - UK & IrelandUsing IESVE for Room Loads Analysis - UK & Ireland
Using IESVE for Room Loads Analysis - UK & Ireland
 
1111 ChatGPT Prompts PDF Free Download - Prompts for ChatGPT
1111 ChatGPT Prompts PDF Free Download - Prompts for ChatGPT1111 ChatGPT Prompts PDF Free Download - Prompts for ChatGPT
1111 ChatGPT Prompts PDF Free Download - Prompts for ChatGPT
 
Structuring Teams and Portfolios for Success
Structuring Teams and Portfolios for SuccessStructuring Teams and Portfolios for Success
Structuring Teams and Portfolios for Success
 

Benchmarking a Scalable and Highly Available Architecture for Virtual Desktops

  • 1. Summary This paper reports on a configuration forVirtual Desktops (VDs) which reduces the total hardware cost to approximately $32.41 per desktop, including the storage infrastructure.This number is achieved using a configuration with dual node, cross-mirrored, High Availability storage. In comparison to previously published reports, which tout the storage infrastructure costs alone ofVDI at from fifty to several hundred dollars per virtual machine, the significance of the data becomes self evident. In this report, storage hardware costs become inconsequential. Even more significantly, the reported configuration achieves this result with 220VDs running on a simple pair of low cost servers.The primary innovation consists in co-locating the redundant storage virtualization systems onto the same hardware platforms as theVDs, thus eliminating the need to amortize high-cost, stand alone storage controllers against manyVirtual Server platforms and thousands ofVDs. Previous publications have reported on configurations which use thousands ofVDs to defray the cost of these controllers. Reading between the lines, it becomes immediately apparent that perVD hardware costs rise very sharply as such configurations are scaled downward.Yet, it is precisely these smaller VDI configurations which are the more important from most practical standpoints. On the other hand, this configuration may also be scaled upwards, in a linear fashion, to thousands ofVirtual Desktops, thus eliminating distended configurations created by the search for artificial “sweet spots” at which costs are optimized. Finally, this configuration uses theVSI benchmark and is based on DataCore’s SANmelody software and the Microsoft Hyper-V virtualization platform.The use of Hyper-V in such benchmarking is unusual as theVMware ESX platform is typically used forVDI sizing. DataCore expects similar results with ESX and will publish those results when they become available. Benchmarking a Scalable and Highly Available Architecture forVirtual Desktops by Ziya Aral & Jonathan Ely - DataCore Software WHITE PAPER
  • 2. The Problem Modern Hypervisors use Storage Area Networks (SANs) to provide the storage infrastructure for networks of virtual machines and hosts. Such shared infrastructures are used to provide portability, provisioning capability, and flexibility to VMs. Moreover, the need for such networks grows proportionally as the numbers of VMs are multiplied. As the number of eggs in “the basket” grows, so too does the need for High Availability configurations to keep them running and to prevent outages which encompass, not one or two, but perhaps dozens of machines. AsVMs proliferate, routine management tasks, such as backup and migration, increasingly depend on the storage management functionality of SANs. Multiply these requirements by a factor of 10 or 100, and we arrive at the synergy betweenVirtual Desktops and SANs. The problem forVirtual Desktop implementations is that SANs are often implemented with large and costly storage controllers and complex external storage networks. While these have the advantage of achieving reasonable scalability, they introduce a very large threshold cost toVDI implementations.To overcome this burden, hardware vendors typically benchmark with one to several thousandVDs. Virtual Desktops are at their introductory stage. Many companies, while understanding the potential benefits of the technology, are introducing pilot programs or attempting to fit initialVDI implementations into existing structures. If the granularity ofVDI implementations is to be in the thousands, then the user is forced to consume, not just the “whole loaf”, but an entire bakery truck full of loaves at one sitting … and this before even knowing whether the bread tastes good. The alternative is equally unappetizing.The user “bites the bullet”, and accepts the high threshold cost and complexity of a full-blown SAN while running far less than the optimal number ofVDs. Now, the per-desktop cost of the implementation becomes much larger than it would have been if the “old scheme” of discrete desktops had remained.This is quite an introduction to a new, “cost-saving” technology… as an increasing number of ever-practical bloggers have noted. DataCore This benchmark is run using the DataCore’s software-based storage virtualization environment on the same platform as the hypervisors hosting theVirtual Desktops. The DataCore software is a full featured implementation of a High-Availability SAN, with all of the functions necessary for aVM hosting infrastructure and unusually high performance characteristics. In addition to that, the software has several characteristics which recommend it for this application. Because the software is portable, it can run as aVM under Hyper-V or ESX and may also run on Windows Server 2008 at OS-level, alongside Hyper-V. While this means that it is a consumer of a common hardware resource, it also means that many of the most important SAN interconnects and storage structures (such as block cache) are implemented locally and virtually – on top of a common backplane. Such a scheme thus gains more from its locality to its clients than it “costs” by virtue of being a consumer for common hardware resources, as will be elaborated in the Performance Discussion section below, In addition, the scheme described simplifies the overall configuration by scaling transparently – each time a newVDI host is added, so are the additional storage resources needed to feed it. Moreover, the scaling occurs in both directions, eliminating what is otherwise a difficult problem in appropriately sizing infrastructure. None of this should be interpreted as a “tossing of the gauntlet” by software based controllers against hardware ones. DataCore often runs in conjunction with such hardware controllers. If the user has a need for such devices, then there is no fault. But, if there is no absolute requirement for such devices at the foundation ofVDI architectures – if neither cost nor capability argues for their initial inclusion – then, hardware controllers serve only to skew theVDI architecture. The Benchmark The benchmark used is theVirtual Session Indexer (VSI) Pro 2.1 from Login Consultants.VSI is becoming the standard for this type of workload. 2 In comparison to the “roll-your-own” benchmarks used by several of the storage vendors, and, even in comparison to trace data,VSI appears to be just a shade more I/O heavy (and especially WRITE-weighted). VSI also creates a working set which is somewhat larger than might be expected from discrete desktops.The explanation for this is thatVSI runs every element of its emulated desktop environment on every individualVD. In addition, application synchronization acrossVDs is expressly minimized. On the whole, this is easy to justify as simple workload normalization, necessary for the creation of a repeatable benchmarking environment.VSI, though, errs on the conservative side and is a bit more challenging than many proprietary workloads. Configuration The configuration used in this benchmark consists of two “white box” server nodes, each with an ASUS Z8PE-D18 Dual LGA motherboard, 2 Intel Xeon E5640 Westmere 2.66GHz Quad Core CPUs and an 80GB SATA boot drive.The total cost is $2114.14 per server, exclusive of DRAM and application storage. In addition, each node is configured with 64GB of DDR3 PC3-10600 DRAM (at a cost of $1056.24) and 4 application drives: 2 SAMSUNG HD103SJ SATA disk drives and 2 Western Digital VelociRaptor WD3000HLFS SATA disk drives ($393.60 for all 4).Total server hardware costs are thus $3564.68 per machine and $7,129.36 for the full configuration. Two SATA drives are configured as a Storage Pool, belonging to SANmelody. SANmelody runs at OS level in this configuration. Whether SANmelody runs at this level, or as aVirtual Machine itself, makes little difference. In this case, the former was chosen for benchmark visibility. The Storage Pool, once configured, is used to host a “golden image” of theVDs. It is then “snapped” using the DataCore Snapshot facility, and each of the writable “snaps” is presented as a boot LUN to the VDs. Experiments were tried to determine if multiple golden images might improve performance. Not only did they not help
  • 3. 3 but the existence of multiple source images actually increased the amount of memory required to accommodate the working set in DataCore’s block cache. Finally, the original image and its snapshots are mirrored to the second server (using DataCore synchronous mirrors over iSCSI) to achieve High Availability. Not only is the configuration thus proof against storage failure but theVDs may be restarted on any other server if a more catastrophic failure should occur.Two SATA drives are configured as a second storage pool to receive mirror traffic from the other node. The configuration described above is implemented with a few simple scripts which then launch 110Virtual Desktops per server and hand the entire configuration off to theVSI benchmark. We live in an era in which the primary characteristics of computing machinery of any class are etched in silicon and are thus nearly identical for any member of that class.This configuration was chosen in order to create a baseline reference at the “lowest common denominator”.The choice of server platforms and their costs will vary by the substitution of everything from specific brand names to optimizations for footprint, power consumption, and certain enterprise “features”. Still, none of these variations will impact performance significantly, and may be seen as simple cost deltas to the reference presented here. Results The configuration below was able to host 220 Desktops using theVSI 2.0.1 benchmark.The disclosure and benchmark configuration report is reproduced in full as an addendum to this paper. Performance Discussion: Co-residency –The principle innovation in this benchmark result is the use of DataCore’s storage virtualization system on the same hardware platforms used to hostVirtual Desktops. While conventional wisdom might suggest that the DataCore software would thus become a competitor for the same scarce resources of the hardware platform, such as memory and CPU, numerous experiments proved the opposite to be true. In each case, such co- located configurations easily outperformed configurations with external storage. The reasons are easy to isolate, retrospectively.TheVirtual Desktop application is not particularly I/O intensive and proves to be easily served by DataCore at the cost of very few CPU cycles.The elimination of the lion’s share of external channel traffic serves to further reduce those demands. In return, block-cache latencies become nearly nonexistent, channel overheads disappear, and I/O latencies are eliminated.The resident SANmelody “pays” for itself, without losing anything in the way of capability or portability. From the standpoint of memory, a similar dynamic also holds.Virtual Desktops provide for relatively small working sets which tend to be READ heavy. Since all READs go to the single source volume, which in turn is the basis for the differential system volume for eachVirtual Desktop, the entire grouping caches very well in a relatively small cache size. Larger caches produce diminishing returns. Finally, one predictable advantage to these types of configurations is that they are inherently “self-tuning”. Each time a group ofVirtual Desktops are added, so too is the storage infrastructure necessary to support them. In contrast to generic SANs which are open ended in their requirements and often organized as independent, and sometimes complex, storage networks, in this case the known requirements of the VDI application may be used to make the storage infrastructure largely transparent. Disks – Given efficient WRITE caching, required disk performance may be significantly minimized. In this benchmark, 4 SATA drives were used, configured into two separate DataCore 2-disk pools.The first pool was used for primary storage and the second was used for mirroring the alternate node.The use of a mix of 7,200 and 10,000 rpm devices is not significant. Any four (and perhaps even 3) 7,200 rpm SATA devices of similar characteristics to the Samsungs produced similar results. The importance of this distinction for us is contained in a separate discussion on creating a baseline configuration. We were loath to use anything but the most standard, easily configured SATA components. SAS, Fibre Channel spindles, fast devices, hybrid disks, and SSDs all have their place in the real world and are well understood to deliver a multiple of ordinary SATA performance. Still, the incorporation of such devices in baseline benchmarking has the same effect as buildingVDI architectures around this or that storage array. Such optimizations may become the dozen different tails attempting to wag an increasingly confused dog. Not only do they make comparisons difficult, but they lead to the subtle intrusion of a specific hardware architecture into the realms in which theVDI requirements themselves should be paramount.
  • 4. 4 Memory – The cost of DRAM comprises a significant portion of the total platform costs in this configuration. Decreasing the size of the memory for eachVD is thus a natural temptation for optimizing the number ofVDs supported by each platform. The difficulty is that the one constant in our industry is the continuously changing density and cost structure of DRAM... a fact which argues for the avoidance of DRAM optimization, rather than arguing for it. This difficulty is compounded by the fact that both Microsoft andVMware offer hypervisor features which allow the “over-loading” of the memory allocated to individual virtual machines. In effect, a kind of global virtual memory is created across “machines”, though the actual implementation mechanisms may differ. Finally, it was discovered during the course of DataCore’sVSI benchmarking thatVDs with very small memory allocations – so small that they would have prevented a discrete PC from booting - nevertheless ran normally in this configuration. On examination, it appeared that the performance of the DataCore block cache created a paging hierarchy which did nothing to prevent “thrashing” but did practically absorb it when a match with working set sizes was achieved. In the end, neither “overloading” nor “fast thrashing” were used in the implementation described here. Not only did this allow for fair comparisons to benchmark data which had already been published, but it avoided optimization for schemes which are largely proprietary and independent of theVDI sizing exercise. In a word, the conventional allocation scheme used here avoided “diversions”. The actual DataCore cache size employed was 4 GBs. While larger caches produced somewhat better results, this was the point at which “diminishing returns” could be clearly established. Subtracting this usage and the memory allocation for Windows Server 2008, the remaining DRAM was simply divided by the number of resident VDs in a conventional fixed allocation. Software Costs – It might be protested that since DataCore has used a software-based StorageVirtualization system in place of a hardware controller, that the cost of the software must be reported in order to create a true “apples-to-apples” comparison.This brings with it its own challenges. Most of the hardware controllers which have been benchmarked are fairly minimalist. Much of the feature set bundled into SANmelody is either not available or it is available at additional cost to the reported configurations. With that qualification noted, the total cost of the software was $3848 per node or $34.98 perVirtual Desktop.This includes both the cost of Microsoft Server 2008 R2 Operating System, which hosts the SANmelody environment and the Hyper-V basedVDs, and DataCore’s SANmelody itself. The fully-burdened cost of the benchmarked configuration, including all of its infrastructural software and hardware, but excluding the OS and applications of theVDs, is thus $67.39 per Desktop. Scaling – As has been discussed, the real problem withVDs has been, not in scaling them up to “thousands” ofVirtual Desktops but in scaling them down to practical configurations. It barely needs mentioning that this must occur without radically spiking costs at the low end and also without forgoing the SAN feature set which assures portability, availability, and data redundancy. Otherwise, the very benefits ofVDI are compromised.The DataCore configuration, described herein, does a good job of maintaining low infrastructure costs at 220 Desktops by creating a paired server approach which simultaneously hosts both theVDs and the storage infrastructure. But, what about scaling upward? How does one arrive at “thousands” of Desktops, if that is what is required, and more importantly, is it possible to arrive at such numbers without wildly varying cost increments and complex reconfiguration? To achieve larger configurations, a DataCore topology known as a “Star” is employed. With the Star, the addition of a third node is used to create a Star center or hub. Additional nodes are then added to the Star. Each of the nodes of the Star acts as before, but they mirror their contents to the hub node rather than to each other. In turn, the hub becomes the single point of control for the entire configuration. With this scheme, each additional server added as a point to the star brings with it its own storage devices and infrastructure as well as itsVDI environment.The scheme is self tuning until the hub is no longer able to accommodate secondary mirror traffic. As an added benefit, most degenerativeVDI phenomena, such as “boot storms”, remain contained in quite small and manageable modules without dominating the entire topology. Simulation and initial experimentation indicates that a simple Star will scale performance (and perVD costs) to several thousand Desktops. DataCore will publish results to bolster these initial findings when they become available.
  • 5. 5 Conclusion and Future Directions This paper reports on an architecture for VDI based on DataCore’s SANmelody, Microsoft’s Server 2008/Hyper-V, and standard server hardware. Login Consultant’sVirtual Session Indexer (VSI) is used as the benchmarking vehicle. On the whole, the benchmarked architecture differs from most of the results published by other storage vendors. Instead of constructing aVDI configuration around a storage controller or pre-existing SAN architecture, it builds the SAN around the VDI servers themselves. In particular, the VDI servers also act as virtualization servers. The resulting locality produces results with as much as ten times the price/performance of other published results, while vastly reducing the number ofVDs (220) at which such economics become available.The whole is accomplished on the simplest, “lowest common denominator” hardware. Nevertheless, the environment appears to be “self-tuning” in that it is resistant to major reevaluation as it scales upward, while it is also modular enough to be resilient in the face of hardware failures, reconfigurations andVDI anomalies, such as “boot storms”. Where to from here? Various optimizations immediately come to mind. In DataCore’s lab, we have successfully increased the density ofVDs by more than 50%, to upwards of 175Virtual desktops, on the same server platforms. Memory tweaks, enhanced hardware, and various tuning “tricks” produce the expected results. Still, increases in costs and complexity produce a system which is little better in price/ performance than the one tested here. Sadly, for a group of “old tuners”, we have had to conclude that the benchmarked configuration is optimized enough to cross the threshold for practicalVDI configurations and that additional low-level optimizations make only small, incremental changes. What about high-level optimizations? Well, perhaps…The co-location of the host OS, the Hypervisor, and the virtualization server – alongside theVirtual Desktops – may provide opportunities for tighter integration. From our vantage point, however, that these remain largely independent today offers up an entirely different class of “optimization”. VDI has one element in common with the other major computing movement of our day: cloud computing. Both technologies promise to aggregate very large numbers of “machines” of the same class… that is, they promise large numbers of platforms, the resource requirements of which are similar and which may be predicted before-hand. Instead of heterogeneous and unknown requirements, the promise is of homogeneous and known ones. In our world, what is “known” is also pre-configurable. This creates the opportunity to present one additional level of virtualization and to “divide and conquer” the one overwhelming fact which immediately strikes anyone attempting to work withVDI: the sheer scale of discretely managing many component machines which are, nevertheless, irritatingly similar. What if, instead of attempting to manage hundreds or thousands of virtual machines discretely, one could divide them into arbitrary groups - sub-units, “virtual datacenters”, or in DataCore parlance, “hives” – and then treat the entire hive as one unit? In the this case, a “hive” would consist of group ofVDs (perhaps 50 or 75 of them), all grouped around a DataCore virtualization server and perhaps other servers needed to provide local services… all implemented as virtual machines, all configured to interact with each other and all without regard to any detailed knowledge about the underlying hardware environment.The entire hive would then be administered, moved, and managed as one unit with the virtualization server acting as a single “port” to the outside world. From the standpoint of the environment, all that would be needed would be a rough approximation of capacity: two “hives” over here and four of them, there… This will be the future direction of our work.
  • 6. © 2011 DataCore Software Corporation. All Rights Reserved. DataCore, the DataCore logo and SANsymphony are trademarks or registered trademarks of DataCore Software Corporation. All other products, services and company names mentioned herein may be trademarks of their respective owners. For additional information, please visit: www.datacore.com or e-mail: info@datacore.com About DataCore Software DataCore Software develops the storage virtualization software needed to get the highest availability, fastest performance and maximum utilization from storage resources in physical and virtual IT environments. It provides the critical 3rd dimension on which the success of server and desktop virtualization projects hinge, regardless of the models and brands of storage devices used. 6 0711 Addendum BenchmarkTesting and Configuration The performance benchmark used to determine the maximum number ofVirtual Desktops in this configuration wasVirtual Session Indexer (VSI) rev 2.1.2 by Login Consultants. LoginVSI creates a simulated user workload (Outlook, Internet Explorer, Excel, etc…) on each of the desktops.The desktops report performance information to a common share folder which is analyzed after the completion of the benchmark run. The simulated user workload is initiated by opening a Remote Desktop session to the virtual desktop. In this case, theVSI Launcher program was used in its default configuration to open the sessions, one desktop every 60 seconds.The user workload runs in a continuous 12 minute loop until all of the desktop sessions have been opened and the last one has run through a loop. After all virtual desktops have run their simulated user workload simultaneously and reported their data to a share folder; the VSI Analysis tool was run.The tool compiles the data in Excel and produces a graph and tables showing response times indicating satisfactory or unsatisfactory user experience. Configuration Information: Virtual Desktops - Microsoft Windows 7 Enterprise x86 – Configured with the defaultVSI optimizations and the optimizations recommended in Virtual Reality Check’s whitepaper “Project VRC: Phase III” http://www.projectvrc.nl/ Benchmark – Login ConsultantsVSI 2.1.2 http://www.loginconsultants.com/ PerVSI recommendation, sevenVSI Launchers were used. All Launchers were running asVMs on a Hyper-V host independent from the hosts running the virtual desktops.The OS of these VMs was Microsoft Server 2008 R2. DataCore SANmelody 3.4 http://www.datacore.com/ SANmelody was run with two bug fixes which will be included in a maintenance update scheduled for this quarter. VSI Benchmark Analysis showing a successful run of 220 desktops.