SUM230 Optimizing storage across FlexCast delivery models
0 likes427 views
The document discusses optimizing storage for virtual desktop infrastructure (VDI) deployments using Citrix FlexCast delivery models. It proposes a "micro-pod" design that leverages local server storage instead of shared storage to improve performance and reduce costs. By balancing storage capacity, performance, and user density on standard 1U servers, the micro-pod approach can support 3,000-4,000 users per rack at a hardware cost of around $200 per user. This modular, localized storage model provides sustainable growth without the need for storage redesigns.
![© 2012 Citrix | Confidential – Do Not Distribute
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](https://image.slidesharecdn.com/cdb6ad57-7064-4747-9c9f-40bb4d2eb242-150527230652-lva1-app6891/85/SUM230-Optimizing-storage-across-FlexCast-delivery-models-21-320.jpg)
![© 2012 Citrix | Confidential – Do Not Distribute
Storage capacity vs. user count - Example
Usable Storage capacity* (GiB) User Count
RAID 6 DISKS 8 DISKS 10 DISKS
0 816 1088 1360
5 with HS 544 816 1088
6 544 816 1088
10 408 544 680
50 544 816 1088
RAID 6 DISKS 8 DISKS 10 DISKS
0 116 155 194
5 with HS 77 116 155
6 77 116 155
10 58 77 97
50 77 116 155
* Array with 146Gb drives
USER_COUNT = TOTAL_DISK ÷ [VM_RAM * (X + 1) + W_CACHE]
W_CACHE = 4GB
VM_RAM = 2GB
PAGE_FILE = 1GB, i.e. X = 50%](https://image.slidesharecdn.com/cdb6ad57-7064-4747-9c9f-40bb4d2eb242-150527230652-lva1-app6891/85/SUM230-Optimizing-storage-across-FlexCast-delivery-models-22-320.jpg)
![© 2012 Citrix | Confidential – Do Not Distribute
Storage performance
RPM IOPS
SSD 5,000+
SAS 15,000 175
SAS 10,000 125
SAS 7,200 75
SAS 5,400 50
IO per Disk
RAID PENALTY
0 1
1 2
5 4
6 6
10 2
50 4
Write Penalties User Workloads
ITEM ~VALUE
IOPS per User 20
Size, KB 4-8
Writes, % 80
Reads, % 20
IOPS = [IOPS per DISK]*[Disk Count]*([% of Reads]+[% of Writes] ÷ [RAID Write Penalty])](https://image.slidesharecdn.com/cdb6ad57-7064-4747-9c9f-40bb4d2eb242-150527230652-lva1-app6891/85/SUM230-Optimizing-storage-across-FlexCast-delivery-models-23-320.jpg)
SUM230 Optimizing storage across FlexCast delivery models
- 1. SUM230: Optimizing storage across FlexCast delivery models Alex Danilychev, PhD Angelo Oddo Sales Engineer Sr. Sales Engineer October 15, 2012
- 2. © 2012 Citrix | Confidential – Do Not Distribute Agenda
- 3. Typical elements of VDI design • Blade servers • Shared storage
- 4. Typical VDI challenges • Hardware cost is 80% storage dependent • Bulk storage purchasing impacts incremental growth • Future scalability demands storage redesign
- 5. © 2012 Citrix | Confidential – Do Not Distribute Shared storage growth and user productivity 1,000 500 User Count Cost, AU100 200 500 User Count Productivity per user AU – arbitrary units
- 6. © 2012 Citrix | Confidential – Do Not Distribute Combined productivity and storage evolution Redesign 1,000 500 Combined Productivity 100 200 Cost, AU Combined Productivity 1,000 500 Cost, AU100 200 ? Alternatives AU – arbitrary units
- 8. © 2012 Citrix | Confidential – Do Not Distribute A pod is a discrete, homogeneous, modular unit of data center components Cisco http://www.cisco.com/en/US/docs/solutions//Enterprise/Data_Center/VMDC/large_pod_design_guide/Large_Pod_Design_Guide.pdf
- 9. © 2012 Citrix | Confidential – Do Not Distribute Redesign Productivity improvements with POD deployment 1,000 500 Combined Productivity 100 200 Cost, AU POD #1 POD #2 POD #3 1,000 500 Combined Productivity 100 200 Cost, AU AU – arbitrary units No redesign
- 10. © 2012 Citrix | Confidential – Do Not Distribute Common POD Characteristics • Addresses performance uncertainties through resource segmentation • Not cost effective at low utilization • Targets large implementations • Cost prohibitive for small deployments • Hardware cost per user is $1,000 or higher
- 11. Can we do better?
- 12. © 2012 Citrix | Confidential – Do Not Distribute Citrix FlexCast VM delivery options • Single-tenant VMs: ○ VDI-in-a-Box ○ MCS (Machine Creation Services) – XenDesktop • Single-tenant and multi-tenant VMs: ○ PVS (Provisioning Services) – XenDesktop and XenApp
- 13. © 2012 Citrix | Confidential – Do Not Distribute Write-cache design with shared storage Provisioning Servers Web Interface VMs Desktop Delivery Controller VMs Licensing Server VM XenApp Server VMs XenDesktop VM Instances SQL Server VM Virtual Machines Active Directory Users Provisioned VMs HYPERVISOR SHARED STORAGE
- 14. © 2012 Citrix | Confidential – Do Not Distribute Write-cache design with local storage Provisioning Servers Web Interface VMs Desktop Delivery Controller VMs Licensing Server VM XenApp Server VMs XenDesktop VM Instances SQL Server VM Virtual Machines Active Directory Users Provisioned VMs HYPERVISOR Local write-cache WRITE-CACHE SHARED STORAGE
- 15. micro-POD Design
- 16. © 2012 Citrix | Confidential – Do Not Distribute A micro-pod is a miniature pod leveraging local storage SUM230
- 17. © 2012 Citrix | Confidential – Do Not Distribute Write-cache on local storage 1,000 500 User Count Cost, AU100 200 50 User Count Productivity per user Within individual server AU – arbitrary units
- 18. © 2012 Citrix | Confidential – Do Not Distribute POD trend Traditional trend Cost-Performance Trends Shared Storage Local Storage 1,000 500 Combined Productivity Cost, AU100 200 1,000 500 Combined Productivity 100 200 Cost, AU micro-POD Performance trend micro-POD trend
- 20. © 2012 Citrix | Confidential – Do Not Distribute Fundamentals of balanced design
- 21. © 2012 Citrix | Confidential – Do Not Distribute 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
- 22. © 2012 Citrix | Confidential – Do Not Distribute Storage capacity vs. user count - Example Usable Storage capacity* (GiB) User Count RAID 6 DISKS 8 DISKS 10 DISKS 0 816 1088 1360 5 with HS 544 816 1088 6 544 816 1088 10 408 544 680 50 544 816 1088 RAID 6 DISKS 8 DISKS 10 DISKS 0 116 155 194 5 with HS 77 116 155 6 77 116 155 10 58 77 97 50 77 116 155 * Array with 146Gb drives USER_COUNT = TOTAL_DISK ÷ [VM_RAM * (X + 1) + W_CACHE] W_CACHE = 4GB VM_RAM = 2GB PAGE_FILE = 1GB, i.e. X = 50%
- 23. © 2012 Citrix | Confidential – Do Not Distribute Storage performance RPM IOPS SSD 5,000+ SAS 15,000 175 SAS 10,000 125 SAS 7,200 75 SAS 5,400 50 IO per Disk RAID PENALTY 0 1 1 2 5 4 6 6 10 2 50 4 Write Penalties User Workloads ITEM ~VALUE IOPS per User 20 Size, KB 4-8 Writes, % 80 Reads, % 20 IOPS = [IOPS per DISK]*[Disk Count]*([% of Reads]+[% of Writes] ÷ [RAID Write Penalty])
- 24. Are mechanical disks affordable but just too slow?
- 25. © 2012 Citrix | Confidential – Do Not Distribute RAID performance for 4K IO workloads Write coalescing, i.e. “derandomizing” IO in action. 3,000-6,000 IOPS from 6-10 disks 4K IO, Random Write 4K IO, Random Read Utilization, % Wait 3,000 IOPS IOMeter load, XenServer 6.0.2 IOSTAT output, RAID5, 8 SAS-15k disks
- 26. © 2012 Citrix | Confidential – Do Not Distribute Availability • For reliability, good performance under failure choose: ○ RAID10 • For RAID with 5 disks or more: ○ RAID6
- 27. © 2012 Citrix | Confidential – Do Not Distribute Storage cost TYPE COST, $ SSD 4,000+ SAS 400 Unit Cost TYPE $ PER MONTH, PER TB SSD 3,000 SAS 300 Cost of Ownership
- 28. Quest for an agile well balanced server
- 29. © 2012 Citrix | Confidential – Do Not Distribute Modern 1U server redundant power high capacity internal storage – 6 to 10 SAS drives 1Gb NICs Hypervisor management remote management space for 10Gb NICs for user and PVS traffic
- 30. © 2012 Citrix | Confidential – Do Not Distribute Modern 1U server • 2 socket design, 12-16 cores • 6-10 15k SAS drives, 146-300Gb each • Storage grade RAID controllers (0.5-2Gb cache)
- 31. © 2012 Citrix | Confidential – Do Not Distribute Sufficient capacity and performance • 0.5-1.5 Tb capacity in RAID 10 (6-10 15k SAS drives, 146Gb or 300Gb) • 3,000-6,000 IOPs available • 80-100 concurrent users consume only 1,600-2,000 IOPs at 20 IOPs per user
- 32. © 2012 Citrix | Confidential – Do Not Distribute Per server hardware cost (including storage) • $10,000 to $16,000 per server • Under $200 per user at 80-100 user density per server
- 33. © 2012 Citrix | Confidential – Do Not Distribute Shared vs. local storage • High Availability • Fault Tolerance • Resource rebalancing • Maintenance
- 35. © 2012 Citrix | Confidential – Do Not Distribute Implementation Building block: ○ 1 rack ○ 2 PVS Servers ○ 2-4 XenServer pools, ~8-16 servers each ○ Local 10Gb interfaces ○ Utilize PVS subnet affinity ○ Estimated user count ~ 3,000 to 4,000
- 36. © 2012 Citrix | Confidential – Do Not Distribute 3,000-4,000 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
- 37. © 2012 Citrix | Confidential – Do Not Distribute Summary • Balanced design is key for affordable VDI • Leverage local storage where possible • Do not rule out mechanical disks • Implement modular design for consistent performance and achieve sustainable growth • $200 per user in hardware costs is feasible
- 38. © 2012 Citrix | Confidential – Do Not Distribute References • http://en.wikipedia.org/wiki/IOPS • http://en.wikipedia.org/wiki/RAID • http://en.wikipedia.org/wiki/Point_of_delivery_(networking)
- 39. © 2012 Citrix | Confidential – Do Not Distribute Before you leave… • Conference surveys are available online at www.citrixsummit.com starting Thursday, 18 October ○ Provide your feedback and pick up a complimentary gift at the registration desk • Download presentations starting Monday, 29 October, from your My Organizer tool located in your My Account