Protect Your Apps!
Tuesday, September 18, 2012
Guest post by Commvault
Here's an operational view into what is happening on VMs when a backup operation kicks off. Most VMware backup solutions that leverage VMware's vStorage APIs for Data Protection (VADP) are going to drive a process that looks very similar to this example. Before the backup operation can begin, the running VM and any applications running therein are quiesced so the backup application can generate an application consistent copy. Many backup applications that use VADP perform this operation, which essentially freezes the VM image while a backup copy is generated.
While the VMs and applications are quiesced, any changes to the application or files in the VM are captured in a scratch space, which essentially stores the changes to data until the backup is complete and the VMs and applications are released back into production. Once released the changes are reconciled back to the frozen image and the VMs and applications proceed on with their workloads.
Here's what's happening graphically. You can see it takes just a minute to snap the VM. VMs remained paused during the Copy Out process, which can take several hours – even if you are leveraging Changed Block Tracking (CBT) and deduplication. Then time is required to reconcile the changes that piled up in the scratch space with the paused VM. Remember, the longer the VM is paused (i.e. the longer the Copy Out process takes), the longer the reconciliation period. So a long Copy Out process has a "double whammy" effect on the total time to run a VM backup.
In this process, VMs are potentially quiesced for an extended period of time, introducing greater risk for failed backups and orphaned VM snaps, meaning risk to your data and applications.
Commvault looked at this emerging problem and saw the key to success is in dramatically collapsing the backup window. We realized that if we could implement a process that minimized the impact of the Copy-Out function we could help customers win the race to optimal data protection and eliminate all those issues around orphaned snaps and failed backups.
This is what our IntelliSnap™ technology is designed to accomplish. IntelliSnap technology leverages vCenter integration to quiesce VMs and apps and then integrates into the native hardware array-based snapshot features available in most storage systems today. This allows us to collapse the initial protection window down from hours to seconds.
Of course, we still need to do a Copy Out process to generate a full backup. The hardware snapshot is valuable for rapid recovery of the entire datastore, but they consume more expensive front end disk space. So we use an off-host process to generate a backup copy of the hardware snap. This means we can release VMs back into production after queiscing for just a few minutes for the initial protection and then run our off-host backup job while the VMs and apps have already resumed normal operations.
This approach is transforming the way customers think about data protection because of the speed, efficiency and reduction of risk inherent in this more streamlined process.