Best Practices in Virtualizing RDS and VDI: THE Virtual Reality CheckRuben Spruijt
@rspruijt
Project: Virtual Reality Check
Ruben Spruijt, PQR
• Introduction
• Phase 1 & 2
• Phase 3 (VDI Deep Dive)
• Phase 4 (AppVirt) - PREVIEW
• Unpublished results
Agenda.
.
www.VirtualRealityCheck.net
.
Performance Analysis & Review
VDI + SBC
.
Independent & Unbiased
.
Latest Gear
.
.
SBC (TS/XenAPP)
Terminal Server
ADFile Share
Hypervisor
Terminal Server
Logging
VDI
Hypervisor
7
ADFile Share
Logging
7 7 7
• Office 2007: •Outlook, •Word, • PowerPoint• Excel
• PDF printer & Adobe PDF
• Internet Explorer (multiple sites + Flash Video)
• FreeMind
VSI workload (14 minute loop)
Phase 1 & 2.
• Virtualization TS/XenApp: Yes we can!• Sweet spot = performance + licensing
• Best-Practices•No vCPU overcommit + 2vCPU minimum•Dedicated Host / Reservation, Enforce vMMU! (x86 - vSphere)
• Intel Nehalem/Westmere: impressive•Hyper-Threading + EPT-D = Doubling Capacity
Conclusions Phase 1 & 2.
Phase 3.
Test platform Phase 3: VMware vSphere 4.xServer Brand/Model HPDL380G6
CPU 2 x Intel Quad core [email protected] ‘Nehalem’ (16 logical cpu!)
Memory 96GB DDR3Disk 8 x 146Gb, 820.2Gb, dual port 10.000RPM Serial SCSI
RAID level RAID-5 with online spareRAID controller HP Smart Array P400i, with 512MB and Battery Backed
Write CacheNIC NC373i Gigabit Adapters, Broadcom 5708
.
VSI vs VRC Optimizations
.
VRC optimizations: Standard VSI Tuning
Win7 default
Win7 VSI opt
Win7 BP opt
Virtual Machine Properties>Options>General>Enable Logging -> Uncheck
V
Aligned disk V V V Disable Offline Files V V Disable Autoupdate V V Disable Background Defrag V V Disable Background Layout Service XP Disable System Restore XP Disable Last Access Timestamp V V Disable Hibernate V V Disable Bug Check Memory Dump V V Disable Indexing Service XP Disable Move to Recycle Bin V V Reduce Event Log Size to 64k V V
.
VRC optimizations: Extra Disabled Services
Disable services Win7 default
Win7 VSI opt
Win7 BP opt
Application Experience V Base Filtering Engine V Background Intelligent Transfer service V Diagnostic Policy service V Function Discovery Resource Publication V Offline files V Superfetch V V TCP/IP NetBios Helper Themes V Windows Defender V V Windows Search V V Windows Update V V Windows Firewall V WLAN Autoconfig V Windows media player Network Sharing Service V Routing and remote Access V HomeGroup Provider V Internet Connection Sharing V Media Center Extender Service V Net.Tcp Port Sharing Service V
.
VRC optimizations: Scheduled Tasks
Disabled Scheduled tasks Win7 default
Win7 VSI opt
Win7 BP opt
\Microsoft\Windows\Defrag\ScheduledDefrag V \Microsoft\Windows\Diagnosis\Scheduled V \Microsoft\Windows\DiskDiagnostic\DataCollector V \Microsoft\Windows\Maintenance\WinSAT V \Microsoft\Windows\Power Efficiency Diagnostics\AnalyzeSystem
V
\Microsoft\Windows\Registry\RegIdleBackup V \Microsoft\Windows\SystemRestore\SR V \Microsoft\Windows\WindowsBackup\ConfigNotification V \Microsoft\Windows Defender\MP Scheduled Scan V \Microsoft\Windows Defender\MPIdleTask V
.
“…Project VRC must emphasize that it is crucial to test and validate these
optimizations in your own VDI deployment…”
VDI + Storage.
• Windows XP
• Windows 7 + VSI Optimizations
• Windows 7 + VRC Optimizations
Understanding I/O
• Boot and Logon
• First VSI loop
• Second VSI loop
• Idle + 6 Apps
• Idle + No Apps
Understanding I/O
• Boot and Logon• First VSI loop
• Second VSI loop
• Idle + 6 Apps
• Idle + No Apps
Understanding I/O
Boot and Logon
Boot and Logon; normalized
Boot and Logon, Read vs Write I/O
• Boot and Logon
• First VSI loop• Second VSI loop
• Idle + 6 Apps
• Idle + No Apps
Understanding I/O
First VSI Loop
First VSI Loop: normalized
First VSI Loop, Read vs Write I/O
• Boot and Logon
• First VSI loop
• Second VSI loop• Idle + 6 Apps
• Idle + No Apps
Understanding I/O.
Second VSI Loop.
Second VSI Loop: normalized.
Second VSI Loop: Read vs Write I/O.
• Boot and Logon
• First VSI loop
• Second VSI loop
• Idle + 6 Apps• Idle + No Apps
Understanding I/O.
Idle I/O With 6 Applications Running.
Idle I/O With 6 Applications Running: Normalized
.
• Boot and Logon
• First VSI loop
• Second VSI loop
• Idle + 6 Apps
• Idle + No Apps
Understanding I/O.
Idle I/O With No Running Apps.
Idle I/O With No Running Apps: Normalized
.
1GB vs 2GB Memory (VRC Opt).
1GB vs 2GB Memory (VRC Opt).
Fixed Or Automatic Page File Size (VRC Opt)
.
IO impact Avirus on VDI
VIRUSSCANNER: TOTAL I/O’s
To be continued…..
Conclusion I/O’s.
• Windows 7 more than Windows XP• Boot• First Loop
• Windows 7 less than Windows XP• Second Loop• Idle
• Best practice: • Use ‘VRC’ Optimizations• fix page file• A/V
Conclusion I/O’s
‘Red Bull’
VMware vSphere:
• HaltingIdleMsecPenalty = HIMP
• Default = 100
‘Red Bull’
VDI on ‘Red Bull’; Windows XP
VDI on ‘Red Bull’; Windows XP
“VDI on Red Bull”; Windows 7
“VDI on Red Bull”; Windows 7
Hyper-Threading Disabled
Hyper-Threading Disabled
• HIMP = VMware vSphere; >> Hyper-V by design fast #LOL
• 23% Improvement Windows XP
• 22% Improvement Windows 7
• Only on Hyper-Threading
• HaltingIdleMsecPenaltyMax
HIMP Conclussion.
Phase 3 - VSI Results.
VSI vs VRC Optimizations.
1 vCPU vs 2 vCPU.
Memory Overcommit (vSphere)Win7 120VM Pre-Booted (96GB RAM)
.
SBC vs VDI
SBC vs VDI
SBC vs VDI
• VSI vs VRC Optimizations
• 1 vs 2vCPU
• Memory overcommit
• SBC vs VDI
VSImax Conclusions
Preview – Phase 4
PREVIEW
.
A-Team
• VSIMax
• Local vs Streaming
• I/O Impact
Phase 4: Impact AppVirt on VDI
• 3 major AppVirt vendors:• Citrix Application streaming•Microsoft App-V•VMware ThinApp
• Office 2007 suite virtualized as 1 package
• Different scenario’s: streamed, precached, shared cache
AppVirt on VDI: test setup
“Project VRC’s goal is to investigate overall performance impact of AppVirt in VDI.”
“Project VRC does not recommend virtualizing the Microsoft Office suite as an overall
best practice.”
• Maximum users on a host
VSIMax results
App-V vs. Local installed %
• Impact VSIMax: ~24%
• Different configurations doesn’t impact VSIMax
App-V conclusions
streamed vs. Local installed %
Bug > Fix!
Pre-cached vs. Local installed %
Bug > Fix!
VDI + Storage - IOPS.
IO reads: stream vs. local %
IO reads: cached vs. local %
IO writes: stream vs. local %
IO writes: pre-cached vs. local %
Worst/worse case scenario or reality?:
• Impact VSIMax: -24%
• Streaming apps = 22-45% less READ IO’s
• Streaming apps = 20-45% more WRITE IO’s
• Check response times in whitepaper (release soon)
• Application Virtualization IS key in Optimized (virtual) Desktop
Appvirt: Conclusion
VDI + Storage - IOPS.
Fusion IO.
Fusion IO.
Fusion IO.
Fusion IO.
Fusion IO.
Finally…..
Hyper-V SP1• Impact of Dynamic Memory in SP1 =
Windows 7 Dynamic Memory (SP1)
Hyper-V SP1• Impact of Dynamic Memory in SP1 =
To be continued…..
• Head-2-Head: VDI on Hyper-V/vSphere/Xenserver (Phase 5)
• AntiVirus impact and optimization in VDI (Phase 6)
• VDI: Remoting Protocols; impact on platform (Phase 7)
• VDI: Storage Optimization/Solutions (TBD)
• .. Ideas!? >> @
VRC vNext 2011.
• www.virtualrealitycheck.net• www.twitter.com/ProjectVRC• VDI + Storage = Deep Impact http://bit.ly/5HTajV
(PQR,@brugh)• Special thanks to: Sven Huisman (PQR, @svenh)
Related Info
.
.
.
Stay up to date with TechNet Belux
Register for our newsletters and stay up to date:http://www.technet-newsletters.be
• Technical updates• Event announcements and registration• Top downloads
Join us on Facebookhttp://www.facebook.com/technetbehttp://www.facebook.com/technetbelux
LinkedIn: http://linkd.in/technetbelux/
Twitter: @technetbelux
Download MSDN/TechNet Desktop Gadget
http://bit.ly/msdntngadget
TechDays 2011 On-Demand
• Watch this session on-demand via TechNet Edge http://technet.microsoft.com/fr-be/edge/
http://technet.microsoft.com/nl-be/edge/• Download to your favorite MP3 or video player• Get access to slides and recommended resources by the speakers