The session I ran on how to design CSV for Hyper-V backups, and how to use DPM 2010, at the Microsoft/System Dynamics Private Cloud Academy in Dublin, Ireland.
8. About Aidan Finn Infrastructure Team Lead at System Dynamics http://www.systemdynamics.ie Working in IT since 1996 MCSE & MVP (Virtual Machine) Experienced with Windows Server/Desktop, System Center, virtualisation, and IT infrastructure. Blog: http://www.aidanfinn.com
10. Also Mastering Windows Server 2008 R2 (Sybex, 2009) - 4 chapters Mastering Windows 7 Deployment (Sybex, 2011) 10215A: Implementing and Managing Microsoft Server 2008 R2 Virtualization – Technical reviewer
11. Agenda Breaking news from MMS 2011 (Level 100) Hyper-V cluster design & backup (Level 400) Introducing DPM 2010 (Level 100) Protecting Hyper-V and workloads (Level 300) Site-to-site replication for DR (Level 300)
13. SCVMM 2012 Beta System Center Virtual Machine Manager 2012 Beta A big leap forward!
14. Fabric Management Hyper-V and cluster lifecycle management – deploy hosts from bare metal and create clusters Third Party Virtualization Platforms - Citrix XenServer and VMware ESX Hosts and Clusters Network Management – Manage IP Address Pools, MAC Address Pools and Load Balancers Storage Management (SMI-S) – Classify storage, Manage Storage Pools and LUNs
15. Resource Management Dynamic Optimization – proactively balance the load of VMs across a cluster Power Optimization – schedule power savings to use the right number of hosts to run your workloads – power the rest off until they are needed. PRO – integrate with System Center Operations Manager to respond to application-level performance monitors.
16. Cloud Management Abstract server, network and storage resources into private clouds Delegate access to private clouds with control of capacity, capabilities and user quotas Enable self-service usage for application administrator to author, deploy, manage and decommission applications in the private cloud
17. Service Management Define service templates Compose operating system images and applications during service deployment Scale out the number of virtual machines in a service Service performance and health monitoring integrated with Ops Mgr Decouple OS image and application updates through image-based servicing. Server application virtualization technologies such as Server App-V
18. Project Concero Portal application Integrated public (Azure) and private (SCVMM) cloud management Still early days
19. DPM 2012 Some details revealed New approach to VSS Item level recovery from VHD, even when DPM in a VM Manage many DPM servers from OpsMgr Role-based administration Can upgrade DPM 2010
21. Why Are We Talking About This? Let’s do some recap And discuss some concepts
22. Backup Strategies There are 3 types of VM backup you can do with Hyper-V (mix ‘n’ match): Treat the VM as a physical machine Backup everything Limited usefulness in Hyper-V world Backup just the changing data Limited backup traffic and more frequent Backup the VM at the physical storage level Replace the “bare metal recovery”
27. Volume Shadow Copy Service (VSS) VSS is used in storage level backup By default the system (software/Windows) VSS provider is used Slow VM backup must be “serialized”
28. Serialization Groups VMs DPM can only backup 1 group at a time Recommended strongly by MS when using system VSS provider for storage level backup “Recovery Point Creation Failed” http://technet.microsoft.com/en-us/library/ff634192.aspx Set registry key value Run DSConfig.ps1 on a protected cluster host Store DataSourceGroups.xml at %PROGRAMFILES%icrosoft DPMPMonfig on DPM server
29. Hardware VSS Provider Try to choose hardware SAN with hardware VSS provider Faster No need to serialize Must have VSS support for CSV
30. Linux & VSS VSS is a feature of Windows No Linux support for VSS Compatible Windows VMs suffer no downtime Two options Stop VM, backup, start VM (requires scripting) Treat the Linux VM as a “physical” server Don’t backup at storage level
31. Reminder: Redirected I/O Exclusive access required to all files for file system operations, such as VSS snapshot. How is that done? CSV coordinator takes over all files. Other host I/O redirected via CSV Coordinator. Also provides storage path fault tolerance.
33. So What?!?!? Backup of 1 VM on CSV causes Redirected I/O for all Need to consider bandwidth and latency for Redirected I/O The storage performance of every VM on a CSV will be impacted What about active/active multi-site clusters? And we haven’t talked about: Performance (IOPS) Storage level replication policies Storage fault tolerance
34. Host Design for CSV All identical OS builds & configurations Almost identical hardware Identical drive letter (C: for %systemdrive% NTLM (NT LAN Manager) must be enabled SMB enabled on each network that may be involved in CSV: Client for Microsoft Networks File and Printer Sharing for Microsoft Networks
35. Host Networking Microsoft recommends host as follows: NIC1: Parent partition NIC2: Virtual machine networking NIC3: Live Migration (private network) NIC4: Cluster Shared Volume (private network) NIC5 & NIC6: iSCSI (for iSCSI SAN only) Some variations HP Flex10 for blades Backup network NIC CSV NIC used for Redirected I/O Label NICs identically on all hosts
36. Specifying CSV NIC CSV will automatically use the private (not routed) network with lowest metric You can force CSV to use a network using PowerShell: CSV NIC labeled as “CSV Network” $n = Get-ClusterNetwork “CSV Network”;$n.Metric = 1033 You can reset to automatic metric with: $n = Get-ClusterNetwork “CSV Network”;$.AutoMetric = $true
37. How Many CSVs And How Big? No right answer Use assessment or performance monitoring data to figure out how many VMs in CSV Use IOPS from assessment VS IOPS capabilities of storage from manufacturer Consider VM placement Clustered or load balanced VMs should be on different storage “All your eggs in one basket”
38. Backup Impact on CSV Design Define backup requirements for VMs Storage level backup once per night/week/month? Data backup? DR replication Maybe start the on-paper design with one CSV Divide into smaller CSVs to limit impact of redirected I/O Backup policy Affinity/fault tolerance Storage level replication Don’t forget IOPS
39. Private Cloud Tough To Design No assessment data Nothing to P2V Nothing to assess You have no empirical data to size from Private cloud is like a Kevin Costner movie “If you build it, they will come” My tip: Build a pilot environment Deploy VMs onto different sized VMs Backup each CSV Measure performance and size accordingly
40. Some Common Approaches to CSV One CSV per host One CSV coordinator role assigned per host Start with one or two CSVs Monitor storage performance with OpsMgr Grow CSVs as required Add CSV before performance degrades Some do scripting to control Redirected I/O: Live migrated CSV VMs to CSV coordinator host Perform CSV backup
41. RAID and VM/VHD Placement RAID 5 gives the most capacity RAID 5 loses maybe 75% of write speed Rule of thumb: If you would deploy RAID 10/5 for a physical box for performance, then place the VM/VHD on RAID 10/5 CSV Complicated by disk group/virtual disk in modern SAN Consult your storage vendor on best performance VS capacity
43. Cluster/CSV Summary Understand requirements for: Backup DR replication Design cluster networking according to best practices Design CSV according to requirements for: Backup policy Performance (IOPS & Redirected I/O) DR storage level replication Fault tolerance
44. Backup Software For storage level backup You can use any backup product certified for Windows Server 2008 R2 Hyper-V Test before you buy/use Cluster VM location awareness CSV support For VM level backup Any backup product should work – check with vendor for support of agent in VM
46. System Center Data Protection Manager 2010 Microsoft’s backup solution Private cloud and Dynamic Datacenter Best of breed for protecting Microsoft products SQL Server, Windows Server, Windows client, Sharepoint, Exchange, Dynamics, and so on License as: Server & agent (Client, Std or Ent) System Center Management Suite Enrolment for Core Infrastructure SCE + DPM + agent
47. Additional DPM 2010 Features Improved storage: auto grow and self healing SharePoint 2010 infrastructure awareness Automatic protection of new SP/SQL resources Self-service Hyper-V CSV support VM, workload, or bare metal backup and restore D2D, D2T, and D2D2T backup PowerShell cmdlets Site-to-site replication AKA DPM2DPM4DR
48. Scalability 100 Servers, 1000 Laptops, up to 2000 Database per Server Up to 80 TB per DPM server
49. DPM Backup Is A Little Different Deploy disk storage/tape library Disk shared by many backup policies Deploy a single all-purpose agent to protected machine Create a protection group (backup policy) Defines what you backup/when/retention policy Agent synchronises with DPM server Block level changes since last backup are replicated, maybe every 15 minutes, via VSS Recovery points are created Maybe 3 times/day. You restore from these.
50. DPM Requirements Check TechNet Too many variations to include here Depends on OS and software Please install required patches D2D storage pools Use “cheap” SATA disk Rule of thumb sizing: 2.5 – 3 times protected disk space Storage calculators available for free download
52. Reminder … We have 3 ways we can backup: Storage/host Treat the VM like a physical machine Backup the changing data We need to associate backup policy with virtual machine CSV placement To control Redirected I/O
53. Scenario 1: Problem Protect 2 VMs File1 (File Server): Backup entire VM once/week Backup file shares 1 time/day SD-SQL1 (SQL Server): Backup entire VM once/day Backup databases 3 times/day Minimise impact on network for backup activity
54. Scenario 1: Solution – Part 1 Place File1 on CSV1 Create protection group for all VMs on CSV1 Backup once per week Install agent on File1 Create protection group: select file shares Synchronise every 15 minutes Restoration point 1 time per day
55. Scenario 1: Solution – Part 2 Place SD-SQL1 on CSV2 Create protection group for all VMs on CSV2 Backup once per day Install agent on SD-SQL1 Create protection group: select all databases Synchronise every 15 minutes Restoration point 3 times per day
57. Disaster Recovery Problem A legal or desired requirement for many organisations Many solutions, including: Point solution replication: complicated and sometimes delicate Storage replication: expensive hardware/licensing/networking and the domain of big corporations
59. Site A Hosts Site B Hosts DPM 2010 DPM 2010 Storage Pools Storage Pools
60. Disaster Recovery Solution What is a virtual machine, really? Just a few files Files are easy to backup and replicate DPM 2010 backs up Hyper-V VMs DPM server in Site A can replicate storage pools to other DPM server in Site B Scheduling and bandwidth throttling Can restore VMs to alternative hosts/clusters in Site B Admin console or by (scheduled) script
61. Scenario 2: Problem Be able to restore File1 and SD-SQL1 to DR site (from Site A to Site B) Recover entire VM and business data to within a few hours Consolidate vendors Minimise costs Hardware, software, and network Minimise complexity
62. Scenario 2: Solution Deploy Hyper-V hosts in Site B Deploy replica DPM server in Site B Configure protection group: Site A DPM database File1 and SD-SQL1 VMs File1 shared folders & SD-SQL1 databases Invocation plan: Recover File1 and SD-SQL1 on Site B hosts Recover File1 shares and SD-SQL1 databases
64. Cross-Platform Protection DPM 2010 protects only Microsoft products i365 have created a DPM-based appliance with addition protected products Physical and virtual appliance Protected products include: VMware RedHat and SUSE Linux Solaris, HP-UX, AIX, IBM i, NetWare Oracle databases
65. Action Contact us for virtualisation assessment 1 free day of consulting for today’s attendees (per organization) Help you determine the right strategy for you Paul.Hall@systemdynamics.ie 01 4830355
66. Private Cloud Academy One of four modules Module I: Hyper-V and Private Cloud Computing Module II: Managing Hyper-V (14 January 2011) Module III: Hyper-V and Data Protection Manager (Today) Module IV: Windows Server 2008 R2 Service Pack 1 (20 May 2011) Very important that we learn from today so please fill out your feedback forms.
67. Did I Stay On Schedule? Sales http://www.systemdynamics.ie Paul.Hall@systemdynamics.ie Aidan Finn Aidan.Finn@systemdynamics.ie @joe_elway http://www.aidanfinn.com