How to build a winning solution for large scale VDI deployments
1 like347 views
The document provides guidelines for implementing successful large-scale Virtual Desktop Infrastructure (VDI) deployments, emphasizing the importance of storage solutions such as all-flash arrays for performance and manageability. Key topics include user types and requirements, VDI lifecycle events, assessment strategies, and the benefits of scale-out architectures. Additionally, it details tools and technologies that enhance VDI environments, highlighting automated management and storage efficiency.
How to build a winning solution for large scale VDI deployments
- 1. How to Build a Winning Solution for Large-Scale VDI Deployments
- 2. Your Webinar Presenters Garrett Clark VDI Architect SolidFire Tristan Todd End User Computing Architect VMware, Inc.
- 3. VMware Perspective: How to build a winning solution for large-scale VDI deployments Tristan Todd End User Computing, VMware
- 4. Hosted RDS Desktops & Applications VMware Core Infrastructure Virtual Desktop Pools Windows Full Clone Windows Linked Clone Linux Desktops MIRAGE FLEX SaaS vIDM vApp VMware Horizon View Identity Manager App Volumes UEM Horizon Clients Virtual SAN NSX vCenter vSpher e vRealize Ops View Security Servers View Connection Servers View Composer Servers
- 5. Data Diversity Example – Key VDI Lifecycle Events • Initial Deployment • Initial Login • Desktop in use • Desktop Idle • Single app deploy • Pool Update (recompose) • Pool Update Rollback • Single desktop repair / redeploy • Pool Scale-out • Duplicate Pools (same master/parent) • Desktop reclamation (user remove) • Pool decommission • BCDR failover / recovery • Login Storm • Logoff Storm • Security Definition Updates • Security Scans • Persona Initial Synch • Persona Scheduled Sync • Persona Restore • User installed application install • Background storage operations • Storage Maintenance • Server Maintenance • Network Maintenance • Software Upgrades 5
- 7. Use cases might merit storage segregation 7 User Type Key Requirement Desktop Type Storage Type Websphere Developer Turnover of data Dedicated Full Clone – User Installed Apps Hybrid SAN Graphic Designer Change massive file Dedicated desktop with GPU Local Flash or Flash SAN Admin Assistant Basic Productivity Apps RDSH Desktop or App Candidate Hybrid SAN Finance High IO capability Dedicated or Floating Linked Clone with persona Flash SAN
- 8. What can an All-Flash Array do for you? 8 WebSphere Developer Graphic Designer Admin Assistant Finance Team IT SysAdmin Offshore Temp Worker
- 9. 9 Don’t Gamble – Do an Assessment!
- 13. It’s as Easy as 1-2-3 to Get Started 13 assessment.vmware.com REGISTER • Fill out a short form for access to your personal dashboard • Cloud-based back-end is automatically provisioned for you1 RUN • Download our simple agent and run it to gather the necessary data from your environment • The dashboard keeps you up to date on the progress of your assessment 2 REPORT • Access personalized reports to review your environment’s readiness and your infrastructure requirements • Gain visibility into the costs associated with your Horizon environment (coming soon) 3
- 15. vRealize Operations for Horizon 15 Aggregates metrics into workload, capacity and health scores Relies on dynamic thresholds
- 16. vRealize Log Insight and Device Latency 16
- 18. Liquidware Labs Stratusphere FIT 18
- 19. The thing about IOPs
- 20. Virtual Disk IO – It’s Mostly about Latency Application Guest OS ESX Storage Stack VMM Driver 3. KAVG 2. DAVG 1. GAVG QAVG GAVG = KAVG + DAVG Fabric vSCSI HBA Time in ESXi storage stack is minimal, for all practical purposes KAVG ~= QAVG. In a well configured system QAVG should be zero Array SP When KAVG > 0, I/O is typically backed up in a device or adapter Queue 20
- 21. Our VDI users deserve the best storage 21
- 22. How do we define the “best storage”? Availability Manageability Performance Simplicity Scalability
- 23. Comparing the Categories 23 All Flash Arrays Hybrid Arrays Converged Appliances & “Blocks” Virtual Storage Appliances DAS Virtual SAN Scalability Good Good Good Good Limited Good Performance Excellent Good Good Good Best Amazing Availability Best Best Variable Good LOL Better Manageability Excellent Good Mixed Bag MEH Complex Better Simplicity Best Good Good Good Mixed Bag Good
- 24. SolidFire for VDI • Cost – Global Efficiencies • Scale-Out Architecture – Grow from Pilot to Production to Scale – First VM same performance as the last • Performance Control – Start VDI Pilot for “Free” – Multiple VDI projects • Automated Architecture – Automated HA – Automated Management – Simple
- 25. ! Compression – 2-4x ! De-Duplication VM’s% 15%% 10%% 5%% 2.5%% 50# 6.0x# 8.5x# 14.5x# *# 100# 6.3x# 9.2x# 16.8x# *# 500# 6.6x# 9.8x# 19.3x# 37.1x# 1000# *# 9.9x# 19.6x# 38.5x# 5000# *# 10x# 19.9x# 39.7x# Efficiencies in VDI
- 26. ! Storage Cost, $/Desktop VM’s% 15%% 10%% 5%% 2.5%% 50# $84# $59# $35# *# 100# $80# $55# $30# *# 500# $76# $51# $26# $14# 1000# *# $51# $26# $13# 5000# *# $50# $25# $12# Efficiencies in VDI
- 27. ! New nodes are added as demand dictates ! Performance and capacity instantly available to all volumes ! Nodes added on the fly without down time 34.4 TB 43 TB 51.6 TB 200,000 IOPS 250,000 IOPS 300,000 IOPS 350,000 IOPS Performance Capacity Linear ScalabilityA 4 node SF2405 configuration 4, SF2405 Nodes ! 25.8 TB ! 200,000 IOPS 25.8 TB Scale-out Architecture
- 28. Scaling a SolidFire System ! New nodes can be added or removed dynamically and non-disruptively ! New capacity and performance immediately available ! No manual intervention needed ! Data is automatically redistributed ! Performance is automatically load balanced ! Mix and Match node types ! Make decisions based upon capacity or performance ! No SSD vendor lock-in ! Remove nodes from the system HB D B C F# E# A G D GA HC E# SF2405 ------------------------- ! 8.6 TB effective ! 50k IOPS SF4805 ------------------------- ! 17 TB effective ! 50k IOPS SF9605 ------------------------- ! 34.5 TB effective ! 50k IOPS F#
- 29. Cost Benefit of Scale-Out vs. Scale-Up • Example Environment – 1k desktop per year growth " 4k in 4yrs – Assuming a $50/desktop Cost • Scale-Up – Purchase storage for 4,000 upfront: $200k – “Annual” cost of $50k • Scale-Out – Annual purchase of storage for 1,000 desktops – Take advantage of ~20%/yr price decrease – Total Cost is $148k ($37k annually) – Savings of $53k (>25%) over 4 years Yr#0# Yr#1# Yr#2# Yr#3# Scale*Up# Scale*Out# Savings#
- 32. Eliminate the Need for a Dedicated Array • Best Practice Today is to Dedicate an Array for VDI – VDI performance is very unpredictable/inconsistent • On array purchased for other applications – Use additional capacity for VDI pilot, “VDI for Free” • Purchase the array for VDI – Don’t know the De-duplication rates – Spec Conservatively • Likely using a fraction of the capacity • Get the additional capacity for “Free” – Spec based upon an “Average” • Mitigate risk with Scale-Out
- 33. Data redundancy across entire system ! Protects against any drive failure ! Any controller failure ! Any node (shelf) failure ! Any network failure ! Rebuilds Automatically ! Automated Recoveries Shared Nothing HA ! Protects against 1 or 2 disk failures ! A single controller failure ! NO protection against shelf failure ! Against sequential controller failures ! Against sequential drive failures ! Fire-Drill Recoveries Data redundancy within a single shelf Controller 1 Controller 2 Shared Disk Shelf HA Shared Nothing High Availability
- 34. ! Distributed “Shared Nothing” data protection ! Redundancies across nodes ! Self Healing ! No Manual intervention ! Automatic and complete restoration of redundancy after failure B DC F# E# G D A H C H B A E# F# G 0# 6# 12# 18# 24# Hours#to#Start#the#Rebuild# Hours% Self*Healing# 4#Hour#Support# NBD#Support# SolidFire HelixTM Automated HA
- 35. ! Distributed “Shared Nothing” data protection ! Redundancies across nodes ! Self Healing ! No Manual intervention ! Automatic and complete restoration of redundancy after failure ! Industry leading rebuild rates ! <60 minutes for node ! Rates increase as it scales SolidFire HelixTM Automated HA
- 36. ! Distributed “Shared Nothing” data protection ! Redundancies across nodes ! Self Healing ! No Manual intervention ! Automatic and complete restoration of redundancy after failure ! Industry leading rebuild rates ! <60 minutes for node ! Rates increase as it scales ! Eliminates the need for “fire-drill” recoveries 0# 4# 8# 12# 16# SF#5#node# SF#10#node# All#SSD# Tiered# Hours% Time#to#Rebuild# Time#to#Start# SolidFire HelixTM Automated HA
- 37. TIME YOUR BUSINESS DESIGN IniUal#size#(GB#or#IOPS)# INSTALLATION OPERATING GB#and#IOPS#required# Account#CreaUon# SCALING GB#and#IOPS#needed# UPGRADEDESIGN IniUal#size#(GB#or#IOPS)# AnUcipated#growth## over#lifecycle#of#array# IOPS#required#lifeUme#of#product# Node#size#selecUon# RAID%type,%RAID%grouping% Drive%type%(FC,%NL,%Flash)% Drive#quanUUes#and#raUo# Cache#requirements# Drive#Sparing# Aggregate/CPG# MetaLUN#and#Bin#file#layout# INSTALLATION Bin#File/RAID#Layout# Aggregate/MetaLUN#creaUon# Snapshot%spare%spacing% Vol0/quorum#drive#layout# Aggregate#snapshot#frequency# Snapshot#growth#limits# Data%MigraHon%Strategy% OPERATING GB#and#IOPS#required# Account#CreaUon# Host#to#HBA#assignment## and#tracking# Host#iniUator#count#limits# RAID#type#required# Primary#Control#Node# assignment# IOPS#limitaUon#concerns# Snapshot#growth#policy# Thin#Provisioning#(or#not)# Aggregate%assignment% 40%#CPU#limit#consideraUons# SCALING GB#and#IOPS#needed# Drive%Shelves%required% Power/floor#space#restricUons# Control#node#limitaUons## (CPU#storage)# Controller%Upgrade%or% Replacement% UPGRADE HBA%compaHbility%list% CPU#uUlizaUon#on#control#nodes# Maintenance#window# MiUgate#risk#during#node#outage# Storage#VirtualizaUon# CompaUbility# OTHER VENDORS Traditional Storage System Tasks
- 38. TIME YOUR BUSINESS DESIGN IniUal#size#(GB#or#IOPS)# INSTALLATION OPERATING GB#and#IOPS#required# Account#CreaUon# SCALING GB#and#IOPS#needed# UPGRADEDESIGN IniUal#size#(GB#or#IOPS)# INSTALLATION OPERATING GB#and#IOPS#required# Account#CreaUon# SCALING GB#and#IOPS#needed# UPGRADE Internally Automated Storage System Tasks SOLIDFIRE
- 39. Fuel the Software-Defined Data Center with SolidFire Virtualiza)on, Pla.orm Business,, Value INTEGRATION SOLIDFIRE AUTOMATION BUSINESS VALUE Automated## Management# Guaranteed# Performance# Global## Efficiency# Scale*Out# Data## Assurance# vCenter Plug-in PowerShell Storage Policy Based Management (SPBM) End User Computing Infrastructure Delivery Application Deployment SIOC + QoS VAAI VASA VM% SRM VVol vRealize Suite
- 40. SolidFire for VDI • Cost – Global De-Duplication • Scale-Out Architecture – Grow from Pilot to Production to Scale – First VM same performance as the last • Performance Control – Start VDI Pilot for “Free” • Automated Architecture – Automated HA – Automated Management – Simple
- 41. Thank You