5. Agenda
• Understanding your data
• Defining backup requirements
• Overview of Eduserv’s data and requirements
• Problems with traditional backups in a virtual
datacentre
• Solutions to traditional backup issues
6. Terminology
• Recovery Point objective
• Oldest point the data stored on backups can be
• Recovery Time Objective
• Time allowed to restore the data
• Backup window
• Time which the backup window must complete in
7. Terminology
• Backup/Archive/DR
• Backup: used to recover data following loss/corruption
• Achieve: used to store data long term
• Disaster recovery: policy and process to provide service
continuation in event of catastrophic failure
8. Terminology
• Consistency
• Crash consistent: does not provide guarantees of data
integrity
• File system consistency: guarantees file system state
• Application consistency: guarantees application
consistency
15. Understand your data
• Data structure
• Highly transactional/static content
• Large or small files
• Rate of duplication
• Data use
• Useful life of the data
• Does the data need to be backed
• Who controls the data
16. Requirements
• Why is the data backed up
• Recovery Time Objective
• Recovery Point Objective
• Retention period
• Offsite requirements
• Cost
17. Virtual server data
• Large and small files
• High percentage of duplicated data
• Data change rate varies
• Typically short data life
• What we don’t backup
• We don’t control the data
18. Virtual server backup requirements
• Restores: recover from deletion/corruption
• Backup service that is independent from the OS
• Fast backup and restores with low overhead
• Short retention period
• Ability to restore entire VMs or individual files
• Single site/dual site
• Self service backups
• Scalability
• Low cost
19. A very…brief backup history
• One to one relationship between servers and backup
disks
http://www.flickr.com/photos/sylvar/31436967/sizes/l/in/photostream/
20. A very…brief backup history
• Central backup tape repository for backup storage
• Accessed via a client side agent
• Traditional approach was used initially for virtual
machine backups
21. Problems with traditional backups in a virtual world
• Processing
• High consolidation ratios mean higher impact
• Agent based backups require client resources
• Streaming to tape
• No parallelisation
• High latency
• Issues with long term incremental backups
• High administrative overhead
22. Problems with traditional backups in a virtual world
• Restore time
• Slow to locate and load tapes, and to locate data on tape
• Slow to restore entire VM as process is the same as
physical server
• Storage footprint
• Large storage foot print required as de-dup etc. cannot
easily be used
• To improve restore time full backups taken weekly
23. Resolutions to issues – disk to disk
• Enhanced parallelisation of jobs
• Reduced administrative overhead
• Improved restore time
• Reduced foot print
24. Resolution – move backups to the hypervisor
• No more agents :-)
• Change block tracking
26. Resolution – move backups to the hypervisor
• No more agents :-)
• Change block tracking
• Single backup to provide file level and image level
restore
28. Resolution – move backups to the hypervisor
• No more agents :-)
• Change block tracking
• Single backup to provide file level and image level
restore
• Forever/Reversed incremental
31. Resolution – move backups to the hypervisor
• No more agents :-)
• Change block tracking
• Single backup to provide file level and image level
restore
• Forever/Reversed incremental
• Scale-out infrastructure
33. What does this mean?
• 170GB machine with static data
• Traditional backup: ~3 hours
• Virtualised backup: ~2 minutes
• ~230TB VM data
• Virtualised backup: ~3 hour backup window
• Continue meeting the backup window with horizontal
scaling
34. Conclusion
• Understand your data
• Understand your requirements
• For virtual backups
• Look at disk to disk for virtual platforms
• Make sure you take advantage of low processing
overhead such as change block tracking and single
backups for image and file level restores
• Only backup what you need to!
35. Thank you – questions?
Charles Llewellyn and Matt Johnson