SlideShare a Scribd company logo
1 of 44
Download to read offline
SYN256
Storage I/O and capacity analysis
for affordable VDI
Alex Danilychev, PhD
Sr. Sales Engineer
Citrix Synergy 2014
© 2014 Citrix. Confidential.2
Tweet about this session with
#SYN256 and #CitrixSynergy!
@TEKNICA
© 2014 Citrix. Confidential.3
 Reality check
© 2014 Citrix. Confidential.4
POC
Just getting started
© 2014 Citrix. Confidential.5
Pilot
Yes, we can do VDI!
© 2014 Citrix. Confidential.6
Production
How did we get here?
© 2014 Citrix. Confidential.7
Back to fundamentals
Creating a balanced solution
© 2014 Citrix. Confidential.8
 Fundamentals of
balanced design
© 2014 Citrix. Confidential.9
 Good place to start:
Storage*
* 60-80% of VDI hardware cost is storage dependent
© 2014 Citrix. Confidential.10
 Storage I/O and capacity
 Addressing user experience
© 2014 Citrix. Confidential.11
Looking into VM and user workloads
What is acceptable cost and performance?
PER USER SIZING
RAM 0.5 – 8 GB
CPU slice 1 – 1/15
Network 10 – 1,000 Mb/s
Disk 1 – 20 GB
IOPS 5 – 200
$100 to over $5,000 per user, what is your number?
© 2014 Citrix. Confidential.12
VM lifecycle analysis
Realistic approach, matching business requirements with available budget
Provision Boot Logon Steady state Logoff Shutdown Decommission
© 2014 Citrix. Confidential.13
Data collection and
analysis
Getting the right tools
© 2014 Citrix. Confidential.14
Data collection and analysis
 Data acquisition tools
 Load generators
 Data analysis
•  Perfmon for Microsoft Hyper-V http://goo.gl/dqIXhP
•  ESXTop for VMware ESX/ESXi http://goo.gl/w1pSi
•  IOSTAT for Citrix XenServer http://goo.gl/W3b0v2
•  Actual VM & user load http://www.yourcompany.com
•  ATTO bench32 http://www.attotech.com
•  Iometer http://www.iometer.org
•  Microsoft Excel
© 2014 Citrix. Confidential.15
 Profile storage I/O
 ATTO benchmark tool
© 2014 Citrix. Confidential.16
Storage I/O Profile: Example #1 (high max throughput)
0
100
200
300
400
500
600
Transfersize
Time, sec
The average size of the requests (in sectors)
ATTO bench32
VDI user IO range
VDI infrastructure
related IO range
© 2014 Citrix. Confidential.17
Storage I/O Profile: Example #2 (low max throughput)
0
10
20
30
40
50
60
70
Transfersize
Time, sec
The average size of the requests (in sectors)
VDI user IO range
VDI infrastructure related IO range
ATTO bench32
© 2014 Citrix. Confidential.18
 Profile storage throughput
 Iometer tool
© 2014 Citrix. Confidential.19
Iometer: synthetic “workstation-like” workload
•  Transfer request size: 4KB
•  Write: 80%
•  Read: 20%
•  Random access: 75%
•  Worker threads: 1 to 32
© 2014 Citrix. Confidential.20
Iometer throughput under workstation workload
0
10
20
30
40
50
60
70
80
MBps
Time, sec
Read and Write (in MBps)
rsec/s wsec/s
1 2 4 8 16 32
worker-threads
© 2014 Citrix. Confidential.21
Iometer IOPS under workstation workload
0
2000
4000
6000
8000
10000
12000
14000
16000
18000
I/Opersecond
Time, sec
Read and Write (in I/O per second)
r/s w/s
1 2 4 8 16 32
worker-threads
© 2014 Citrix. Confidential.22
 Continue with VM lifecycle analysis
 Assuming no obvious flaws are found, otherwise get something else
© 2014 Citrix. Confidential.23
VM lifecycle stage example: VM boot
Provision Boot Logon Steady state Logoff Shutdown Decommission
© 2014 Citrix. Confidential.24
I/O per second
0
5000
10000
15000
20000
25000
30000
35000
I/Opersecond
Time, sec
Read and Write (in I/O per second)
r/s w/s
*Citrix Machine Creation Services example (non-persistent pooled VMs)
© 2014 Citrix. Confidential.25
Read and Write Throughput, MBps
0
200
400
600
800
1000
1200
MBps
Time, sec
Read and Write (in MBps)
rsec/s wsec/s
*Citrix Machine Creation Services example (non-persistent pooled VMs)
© 2014 Citrix. Confidential.26
I/O vs Throughput
*Citrix Machine Creation Services example (non-persistent pooled VMs)
0
200
400
600
800
1000
1200
I/Opersecond
Read and Write (in MBps)
IOPS vs Read and Write MB transfer
rsec/s wsec/s
8K-12K IOPS
40K IOPS56K IOPS
© 2014 Citrix. Confidential.27
IO distribution
*Citrix Machine Creation Services example (non-persistent pooled VMs)
1
10
100
1000
0 4 8 12 16 20 24 28 32 36 40 44 48 52 56 60 64 68 72 76 80 84 88 92
Read IO Write IO
IOPSfrequency
12K IOPS
48K IOPS
© 2014 Citrix. Confidential.28
Data transfer
0
5000
10000
15000
20000
25000
30000
35000
TotalMBtransfer
Time, sec
Read and Write MB transfer
reads writes
40%
60%6min boot, 100VMs
*Citrix Machine Creation Services example (non-persistent pooled VMs)
© 2014 Citrix. Confidential.29
Storage Queuing and Latency
 Queuing  Latency, ms
0
1
2
3
4
5
6
7
8
Milliseconds
Time, sec
The average time for I/O requests issued (in milliseconds)
await
0
20
40
60
80
100
120
140
160
Queuelength
Time, sec
The average queue length of the requests
avgqu-sz
*Citrix Machine Creation Services example (non-persistent pooled VMs)
© 2014 Citrix. Confidential.30
Continue with VM lifecycle stage analysis
Provision Boot Logon Steady state Logoff Shutdown Decommission
Go back, if necessary - this is an iterative process
© 2014 Citrix. Confidential.31
 Capacity planning
© 2014 Citrix. Confidential.32
Storage capacity vs. user count*
USER_COUNT * VM_DISK + SWAP = TOTAL_DISK
USER_COUNT * (PAGE_FILE + W_CACHE) + SWAP = TOTAL_DISK
PAGE_FILE = VM_RAM * X and SWAP ~ VM_RAM * USER_COUNT
USER_COUNT = TOTAL_DISK / [VM_RAM * (X + 1) + W_CACHE]
VM_DISK SWAPW_CACHEPAGE_FILE
TOTAL_DISK
*Citrix Provisioning Services example (write cache on the target machine)
© 2014 Citrix. Confidential.33
Storage capacity vs. user count – Windows 7/8
Usable Storage capacity* (TiB) User Count
RAID 6 DISKS 8 DISKS 10 DISKS
0 1.676 2.235 2.794
5 with HS 1.118 1.676 2.235
6 1.118 1.676 2.235
10 0.838 1.118 1.397
50 1.118 1.676 2.235
RAID 6 DISKS 8 DISKS 10 DISKS
0 139 186 233
5 with HS 93 139 186
6 93 139 186
10 70 93 117
50 93 139 186
* Array with 300GB drives
USER_COUNT = TOTAL_DISK / [VM_RAM * (X + 1) + W_CACHE]
W_CACHE = 6GiB
VM_RAM = 4GiB
PAGE_FILE = 2GiB, i.e. X = 50%
*Citrix Provisioning Services example (write cache on the target machine)
© 2014 Citrix. Confidential.34
“”
 SYN256
 “Avoid hidden bottlenecks through
 capacity and workload segmentation”
© 2014 Citrix. Confidential.35
 “POD”, “Block”, etc. approach
 Address performance and capacity through
workload segmentation
© 2014 Citrix. Confidential.36
Common POD/Block characteristics
 Addresses performance and capacity uncertainties
 Not cost-effective at low utilization
 Targets large deployments
 Cost prohibitive for small implementations
 Hardware cost per user could be $800 or higher
© 2014 Citrix. Confidential.37
 Making VDI affordable
© 2014 Citrix. Confidential.38
“”
 SYN256
 “Leverage local storage where possible”
© 2014 Citrix. Confidential.39
Leveraging local storage
MicroPOD approach
 Commodity Hardware: DELL R620, HP DL360 or similar
•  192 - 384 GB RAM
•  10 SAS-15k drives, 300 GB each
 Predictable performance and capacity
•  Usable capacity: 1.5-2TiB
•  Low (<1 ms) storage latency
•  4-8K IO as high as 20,000 per server
 Hardware cost per user is below $150-$200
© 2014 Citrix. Confidential.40
 Putting it all together
© 2014 Citrix. Confidential.41
3,000-4,000 XenDesktop users per standard rack
HA design with no shared storage
VM hosts with
user workloads
Localized PVS traffic
PVS fail-over traffic
PVS servers
Network switch
© 2014 Citrix. Confidential.42
Summary
Balanced design is key for affordable VDI
 Implement modular design for consistent
performance and achieve sustainable
growth
 Leverage local storage where possible
 Do not rule out mechanical disks
 $200 per user in hardware costs is feasible
© 2014 Citrix. Confidential.43
Before you leave…
 Recommended related breakout sessions
•  SYN221
•  SYN315
 Conference surveys are available online
at www.citrixsynergy.com starting
Thursday, May 8 at 9:00 a.m.
•  Provide your feedback by 6:00 p.m. that day to
be entered to win one of many prizes
 Download presentations starting Monday,
May 19 from the My Event Planning tool
© 2014 Citrix. Confidential.44
WORK BETTER. LIVE BETTER.

More Related Content

What's hot

What is this DevOps thing and why do I need it?
What is this DevOps thing and why do I need it?What is this DevOps thing and why do I need it?
What is this DevOps thing and why do I need it?
Safe Swiss Cloud
 

What's hot (20)

JetStor NAS 724UX and 724UX 10G ZFS appliance
JetStor NAS 724UX and 724UX 10G ZFS applianceJetStor NAS 724UX and 724UX 10G ZFS appliance
JetStor NAS 724UX and 724UX 10G ZFS appliance
 
What is this DevOps thing and why do I need it?
What is this DevOps thing and why do I need it?What is this DevOps thing and why do I need it?
What is this DevOps thing and why do I need it?
 
Virtualization for competitive advantage - Eric Vanderburg
Virtualization for competitive advantage - Eric VanderburgVirtualization for competitive advantage - Eric Vanderburg
Virtualization for competitive advantage - Eric Vanderburg
 
ContainerDays NYC 2016: "Containers in Azure: Understanding the Microsoft Con...
ContainerDays NYC 2016: "Containers in Azure: Understanding the Microsoft Con...ContainerDays NYC 2016: "Containers in Azure: Understanding the Microsoft Con...
ContainerDays NYC 2016: "Containers in Azure: Understanding the Microsoft Con...
 
IXcloud, Global 1st Cloud OS, Openstack Cloud
IXcloud, Global 1st Cloud OS, Openstack CloudIXcloud, Global 1st Cloud OS, Openstack Cloud
IXcloud, Global 1st Cloud OS, Openstack Cloud
 
Safe Swiss Cloud: Swiss Enterprise Cloud since 2012
Safe Swiss Cloud: Swiss Enterprise Cloud since 2012Safe Swiss Cloud: Swiss Enterprise Cloud since 2012
Safe Swiss Cloud: Swiss Enterprise Cloud since 2012
 
Webinar: How to Put an End to Hyperconverged Silos
Webinar: How to Put an End to Hyperconverged SilosWebinar: How to Put an End to Hyperconverged Silos
Webinar: How to Put an End to Hyperconverged Silos
 
SIPfoundry CoLab 2013 - Cloud virtualization of the sipXecs communications so...
SIPfoundry CoLab 2013 - Cloud virtualization of the sipXecs communications so...SIPfoundry CoLab 2013 - Cloud virtualization of the sipXecs communications so...
SIPfoundry CoLab 2013 - Cloud virtualization of the sipXecs communications so...
 
Azure en Nutanix: your journey to the hybrid cloud
Azure en Nutanix: your journey to the hybrid cloudAzure en Nutanix: your journey to the hybrid cloud
Azure en Nutanix: your journey to the hybrid cloud
 
Tintricity on the Road: From Virtualization to Cloud
Tintricity on the Road: From Virtualization to CloudTintricity on the Road: From Virtualization to Cloud
Tintricity on the Road: From Virtualization to Cloud
 
Enterprise Grade Cloud Computing
Enterprise Grade Cloud ComputingEnterprise Grade Cloud Computing
Enterprise Grade Cloud Computing
 
Databases love nutanix
Databases love nutanixDatabases love nutanix
Databases love nutanix
 
Getting it Right: OpenStack Private Cloud Storage
Getting it Right: OpenStack Private Cloud StorageGetting it Right: OpenStack Private Cloud Storage
Getting it Right: OpenStack Private Cloud Storage
 
Does Hypervisor matter in OpenStack
Does Hypervisor matter in OpenStackDoes Hypervisor matter in OpenStack
Does Hypervisor matter in OpenStack
 
How to deploy a Private Cloud based on WAP and Nutanix
How to deploy a Private Cloud based on WAP and NutanixHow to deploy a Private Cloud based on WAP and Nutanix
How to deploy a Private Cloud based on WAP and Nutanix
 
Open Stack Days israel Keynote 2017
Open Stack Days israel Keynote 2017Open Stack Days israel Keynote 2017
Open Stack Days israel Keynote 2017
 
Compute, Storage & Networking On Demand | Swiss Enterprise Cloud | Up and Run...
Compute, Storage & Networking On Demand | Swiss Enterprise Cloud | Up and Run...Compute, Storage & Networking On Demand | Swiss Enterprise Cloud | Up and Run...
Compute, Storage & Networking On Demand | Swiss Enterprise Cloud | Up and Run...
 
Got Big Data? Splunk on Nutanix
Got Big Data? Splunk on NutanixGot Big Data? Splunk on Nutanix
Got Big Data? Splunk on Nutanix
 
Gain Storage Control with SIOC and Take Performance Control with QoS from Sol...
Gain Storage Control with SIOC and Take Performance Control with QoS from Sol...Gain Storage Control with SIOC and Take Performance Control with QoS from Sol...
Gain Storage Control with SIOC and Take Performance Control with QoS from Sol...
 
ServedBy the Net Products 2017
ServedBy the Net Products 2017ServedBy the Net Products 2017
ServedBy the Net Products 2017
 

Similar to SYN256_2014

SUM230 Optimizing storage across FlexCast delivery models
SUM230 Optimizing storage across FlexCast delivery modelsSUM230 Optimizing storage across FlexCast delivery models
SUM230 Optimizing storage across FlexCast delivery models
Dr. Alex Danilychev
 
Citrix Synergy 2014 - Syn231 Why cloud projects fail
Citrix Synergy 2014 - Syn231 Why cloud projects failCitrix Synergy 2014 - Syn231 Why cloud projects fail
Citrix Synergy 2014 - Syn231 Why cloud projects fail
Citrix
 
Citrix Synergy 2014 - Syn227 Architecting your private cloud infrastructure f...
Citrix Synergy 2014 - Syn227 Architecting your private cloud infrastructure f...Citrix Synergy 2014 - Syn227 Architecting your private cloud infrastructure f...
Citrix Synergy 2014 - Syn227 Architecting your private cloud infrastructure f...
Citrix
 

Similar to SYN256_2014 (20)

SUM230 Optimizing storage across FlexCast delivery models
SUM230 Optimizing storage across FlexCast delivery modelsSUM230 Optimizing storage across FlexCast delivery models
SUM230 Optimizing storage across FlexCast delivery models
 
Citrix Cloud Master Class June 2014
Citrix Cloud Master Class June 2014Citrix Cloud Master Class June 2014
Citrix Cloud Master Class June 2014
 
Citrix Synergy 2014 - Syn231 Why cloud projects fail
Citrix Synergy 2014 - Syn231 Why cloud projects failCitrix Synergy 2014 - Syn231 Why cloud projects fail
Citrix Synergy 2014 - Syn231 Why cloud projects fail
 
Citrix Synergy 2014 - Syn227 Architecting your private cloud infrastructure f...
Citrix Synergy 2014 - Syn227 Architecting your private cloud infrastructure f...Citrix Synergy 2014 - Syn227 Architecting your private cloud infrastructure f...
Citrix Synergy 2014 - Syn227 Architecting your private cloud infrastructure f...
 
Mellanox's Technological Advantage
Mellanox's Technological AdvantageMellanox's Technological Advantage
Mellanox's Technological Advantage
 
Deploying Applications in Today’s Compute, Storage, and Network Infrastructure
Deploying Applications in Today’s Compute, Storage, and Network InfrastructureDeploying Applications in Today’s Compute, Storage, and Network Infrastructure
Deploying Applications in Today’s Compute, Storage, and Network Infrastructure
 
Cisco UCS Invicta accelerated storage
Cisco UCS Invicta accelerated storageCisco UCS Invicta accelerated storage
Cisco UCS Invicta accelerated storage
 
Mellanox Storage Solutions
Mellanox Storage SolutionsMellanox Storage Solutions
Mellanox Storage Solutions
 
Webinar: Don't believe the hype, you don't need dedicated storage for VDI
Webinar: Don't believe the hype, you don't need dedicated storage for VDI Webinar: Don't believe the hype, you don't need dedicated storage for VDI
Webinar: Don't believe the hype, you don't need dedicated storage for VDI
 
Cisco storage networking protect scale-simplify_dec_2016
Cisco storage networking   protect scale-simplify_dec_2016Cisco storage networking   protect scale-simplify_dec_2016
Cisco storage networking protect scale-simplify_dec_2016
 
PROACT SYNC 2013 - Breakout - End to End uitleg over Cisco UCS
PROACT SYNC 2013 - Breakout - End to End uitleg over Cisco UCSPROACT SYNC 2013 - Breakout - End to End uitleg over Cisco UCS
PROACT SYNC 2013 - Breakout - End to End uitleg over Cisco UCS
 
Presentation cloud, the whole offer
Presentation   cloud, the whole offerPresentation   cloud, the whole offer
Presentation cloud, the whole offer
 
Qualcomm Centriq Arm-based Servers for Edge Computing at ONS 2018
Qualcomm Centriq Arm-based Servers for Edge Computing at ONS 2018Qualcomm Centriq Arm-based Servers for Edge Computing at ONS 2018
Qualcomm Centriq Arm-based Servers for Edge Computing at ONS 2018
 
Cisco MDS Main Session EMC World 2015
Cisco MDS Main Session EMC World 2015Cisco MDS Main Session EMC World 2015
Cisco MDS Main Session EMC World 2015
 
Virtualization Acceleration
Virtualization Acceleration Virtualization Acceleration
Virtualization Acceleration
 
2014 08-21 Citrix cloud works with a single management platform technical web...
2014 08-21 Citrix cloud works with a single management platform technical web...2014 08-21 Citrix cloud works with a single management platform technical web...
2014 08-21 Citrix cloud works with a single management platform technical web...
 
Scoping for BMC Discovery (ADDM) Deployment by Traversys Limited
Scoping for BMC Discovery (ADDM) Deployment by Traversys LimitedScoping for BMC Discovery (ADDM) Deployment by Traversys Limited
Scoping for BMC Discovery (ADDM) Deployment by Traversys Limited
 
How to Stop Over Paying Your VMware Taxes
How to Stop Over Paying Your VMware TaxesHow to Stop Over Paying Your VMware Taxes
How to Stop Over Paying Your VMware Taxes
 
Cisco connect montreal 2018 compute v final
Cisco connect montreal 2018   compute v finalCisco connect montreal 2018   compute v final
Cisco connect montreal 2018 compute v final
 
CloudExpo NYC - Citrix Cloud Platforms Best Practices for Architecting Your C...
CloudExpo NYC - Citrix Cloud Platforms Best Practices for Architecting Your C...CloudExpo NYC - Citrix Cloud Platforms Best Practices for Architecting Your C...
CloudExpo NYC - Citrix Cloud Platforms Best Practices for Architecting Your C...
 

SYN256_2014

  • 1. SYN256 Storage I/O and capacity analysis for affordable VDI Alex Danilychev, PhD Sr. Sales Engineer Citrix Synergy 2014
  • 2. © 2014 Citrix. Confidential.2 Tweet about this session with #SYN256 and #CitrixSynergy! @TEKNICA
  • 3. © 2014 Citrix. Confidential.3  Reality check
  • 4. © 2014 Citrix. Confidential.4 POC Just getting started
  • 5. © 2014 Citrix. Confidential.5 Pilot Yes, we can do VDI!
  • 6. © 2014 Citrix. Confidential.6 Production How did we get here?
  • 7. © 2014 Citrix. Confidential.7 Back to fundamentals Creating a balanced solution
  • 8. © 2014 Citrix. Confidential.8  Fundamentals of balanced design
  • 9. © 2014 Citrix. Confidential.9  Good place to start: Storage* * 60-80% of VDI hardware cost is storage dependent
  • 10. © 2014 Citrix. Confidential.10  Storage I/O and capacity  Addressing user experience
  • 11. © 2014 Citrix. Confidential.11 Looking into VM and user workloads What is acceptable cost and performance? PER USER SIZING RAM 0.5 – 8 GB CPU slice 1 – 1/15 Network 10 – 1,000 Mb/s Disk 1 – 20 GB IOPS 5 – 200 $100 to over $5,000 per user, what is your number?
  • 12. © 2014 Citrix. Confidential.12 VM lifecycle analysis Realistic approach, matching business requirements with available budget Provision Boot Logon Steady state Logoff Shutdown Decommission
  • 13. © 2014 Citrix. Confidential.13 Data collection and analysis Getting the right tools
  • 14. © 2014 Citrix. Confidential.14 Data collection and analysis  Data acquisition tools  Load generators  Data analysis •  Perfmon for Microsoft Hyper-V http://goo.gl/dqIXhP •  ESXTop for VMware ESX/ESXi http://goo.gl/w1pSi •  IOSTAT for Citrix XenServer http://goo.gl/W3b0v2 •  Actual VM & user load http://www.yourcompany.com •  ATTO bench32 http://www.attotech.com •  Iometer http://www.iometer.org •  Microsoft Excel
  • 15. © 2014 Citrix. Confidential.15  Profile storage I/O  ATTO benchmark tool
  • 16. © 2014 Citrix. Confidential.16 Storage I/O Profile: Example #1 (high max throughput) 0 100 200 300 400 500 600 Transfersize Time, sec The average size of the requests (in sectors) ATTO bench32 VDI user IO range VDI infrastructure related IO range
  • 17. © 2014 Citrix. Confidential.17 Storage I/O Profile: Example #2 (low max throughput) 0 10 20 30 40 50 60 70 Transfersize Time, sec The average size of the requests (in sectors) VDI user IO range VDI infrastructure related IO range ATTO bench32
  • 18. © 2014 Citrix. Confidential.18  Profile storage throughput  Iometer tool
  • 19. © 2014 Citrix. Confidential.19 Iometer: synthetic “workstation-like” workload •  Transfer request size: 4KB •  Write: 80% •  Read: 20% •  Random access: 75% •  Worker threads: 1 to 32
  • 20. © 2014 Citrix. Confidential.20 Iometer throughput under workstation workload 0 10 20 30 40 50 60 70 80 MBps Time, sec Read and Write (in MBps) rsec/s wsec/s 1 2 4 8 16 32 worker-threads
  • 21. © 2014 Citrix. Confidential.21 Iometer IOPS under workstation workload 0 2000 4000 6000 8000 10000 12000 14000 16000 18000 I/Opersecond Time, sec Read and Write (in I/O per second) r/s w/s 1 2 4 8 16 32 worker-threads
  • 22. © 2014 Citrix. Confidential.22  Continue with VM lifecycle analysis  Assuming no obvious flaws are found, otherwise get something else
  • 23. © 2014 Citrix. Confidential.23 VM lifecycle stage example: VM boot Provision Boot Logon Steady state Logoff Shutdown Decommission
  • 24. © 2014 Citrix. Confidential.24 I/O per second 0 5000 10000 15000 20000 25000 30000 35000 I/Opersecond Time, sec Read and Write (in I/O per second) r/s w/s *Citrix Machine Creation Services example (non-persistent pooled VMs)
  • 25. © 2014 Citrix. Confidential.25 Read and Write Throughput, MBps 0 200 400 600 800 1000 1200 MBps Time, sec Read and Write (in MBps) rsec/s wsec/s *Citrix Machine Creation Services example (non-persistent pooled VMs)
  • 26. © 2014 Citrix. Confidential.26 I/O vs Throughput *Citrix Machine Creation Services example (non-persistent pooled VMs) 0 200 400 600 800 1000 1200 I/Opersecond Read and Write (in MBps) IOPS vs Read and Write MB transfer rsec/s wsec/s 8K-12K IOPS 40K IOPS56K IOPS
  • 27. © 2014 Citrix. Confidential.27 IO distribution *Citrix Machine Creation Services example (non-persistent pooled VMs) 1 10 100 1000 0 4 8 12 16 20 24 28 32 36 40 44 48 52 56 60 64 68 72 76 80 84 88 92 Read IO Write IO IOPSfrequency 12K IOPS 48K IOPS
  • 28. © 2014 Citrix. Confidential.28 Data transfer 0 5000 10000 15000 20000 25000 30000 35000 TotalMBtransfer Time, sec Read and Write MB transfer reads writes 40% 60%6min boot, 100VMs *Citrix Machine Creation Services example (non-persistent pooled VMs)
  • 29. © 2014 Citrix. Confidential.29 Storage Queuing and Latency  Queuing  Latency, ms 0 1 2 3 4 5 6 7 8 Milliseconds Time, sec The average time for I/O requests issued (in milliseconds) await 0 20 40 60 80 100 120 140 160 Queuelength Time, sec The average queue length of the requests avgqu-sz *Citrix Machine Creation Services example (non-persistent pooled VMs)
  • 30. © 2014 Citrix. Confidential.30 Continue with VM lifecycle stage analysis Provision Boot Logon Steady state Logoff Shutdown Decommission Go back, if necessary - this is an iterative process
  • 31. © 2014 Citrix. Confidential.31  Capacity planning
  • 32. © 2014 Citrix. Confidential.32 Storage capacity vs. user count* USER_COUNT * VM_DISK + SWAP = TOTAL_DISK USER_COUNT * (PAGE_FILE + W_CACHE) + SWAP = TOTAL_DISK PAGE_FILE = VM_RAM * X and SWAP ~ VM_RAM * USER_COUNT USER_COUNT = TOTAL_DISK / [VM_RAM * (X + 1) + W_CACHE] VM_DISK SWAPW_CACHEPAGE_FILE TOTAL_DISK *Citrix Provisioning Services example (write cache on the target machine)
  • 33. © 2014 Citrix. Confidential.33 Storage capacity vs. user count – Windows 7/8 Usable Storage capacity* (TiB) User Count RAID 6 DISKS 8 DISKS 10 DISKS 0 1.676 2.235 2.794 5 with HS 1.118 1.676 2.235 6 1.118 1.676 2.235 10 0.838 1.118 1.397 50 1.118 1.676 2.235 RAID 6 DISKS 8 DISKS 10 DISKS 0 139 186 233 5 with HS 93 139 186 6 93 139 186 10 70 93 117 50 93 139 186 * Array with 300GB drives USER_COUNT = TOTAL_DISK / [VM_RAM * (X + 1) + W_CACHE] W_CACHE = 6GiB VM_RAM = 4GiB PAGE_FILE = 2GiB, i.e. X = 50% *Citrix Provisioning Services example (write cache on the target machine)
  • 34. © 2014 Citrix. Confidential.34 “”  SYN256  “Avoid hidden bottlenecks through  capacity and workload segmentation”
  • 35. © 2014 Citrix. Confidential.35  “POD”, “Block”, etc. approach  Address performance and capacity through workload segmentation
  • 36. © 2014 Citrix. Confidential.36 Common POD/Block characteristics  Addresses performance and capacity uncertainties  Not cost-effective at low utilization  Targets large deployments  Cost prohibitive for small implementations  Hardware cost per user could be $800 or higher
  • 37. © 2014 Citrix. Confidential.37  Making VDI affordable
  • 38. © 2014 Citrix. Confidential.38 “”  SYN256  “Leverage local storage where possible”
  • 39. © 2014 Citrix. Confidential.39 Leveraging local storage MicroPOD approach  Commodity Hardware: DELL R620, HP DL360 or similar •  192 - 384 GB RAM •  10 SAS-15k drives, 300 GB each  Predictable performance and capacity •  Usable capacity: 1.5-2TiB •  Low (<1 ms) storage latency •  4-8K IO as high as 20,000 per server  Hardware cost per user is below $150-$200
  • 40. © 2014 Citrix. Confidential.40  Putting it all together
  • 41. © 2014 Citrix. Confidential.41 3,000-4,000 XenDesktop users per standard rack HA design with no shared storage VM hosts with user workloads Localized PVS traffic PVS fail-over traffic PVS servers Network switch
  • 42. © 2014 Citrix. Confidential.42 Summary Balanced design is key for affordable VDI  Implement modular design for consistent performance and achieve sustainable growth  Leverage local storage where possible  Do not rule out mechanical disks  $200 per user in hardware costs is feasible
  • 43. © 2014 Citrix. Confidential.43 Before you leave…  Recommended related breakout sessions •  SYN221 •  SYN315  Conference surveys are available online at www.citrixsynergy.com starting Thursday, May 8 at 9:00 a.m. •  Provide your feedback by 6:00 p.m. that day to be entered to win one of many prizes  Download presentations starting Monday, May 19 from the My Event Planning tool
  • 44. © 2014 Citrix. Confidential.44 WORK BETTER. LIVE BETTER.