For as long as I can remember -- and that’s back to the Jurassic period of computing when mainframes ruled the data center -- system administrators have considered backup and restore to be two sides of the data protection coin. As backup technology has advanced, a new feature -- instant recovery -- has emerged that can eliminate the time-consuming restore stage, allowing administrators to bring their applications back online in minutes.
While we frequently talk about using backups as a key portion of an organization’s disaster recovery plan, the truth is that restores take a long time. When your email or database server has multiple disk failures or has a nervous breakdown and corrupts its database, users aren’t going to be happy when you tell them it will only take a few hours to restore the server to health.
While it can be convenient to blame long restore times on the hours it might take for an incompetent backup administrator to find and mount the right tape -- or more likely to have the tape returned from off-site storage at Recall or Iron Mountain -- tape retrieval isn’t the real problem. Even if the data you need is online, you’ll have time for a cup of coffee or even a five-course meal at a four-star restaurant while that large database is restored.
If -- and that's a bigger if than you think -- your database server can accept data at an LTO-6 tape drive’s rated speed of 160MB/s, restoring a 1TB server will take almost two hours longer if you’re restoring over a 1Gbps Ethernet link. You could get higher throughput from a disk target like a Data Domain or Quantum DXi, but you won’t get anywhere near the data rate their manufacturers advertise. Vendors advertise the aggregate rate at which their systems can ingest data when accepting multiple backup streams in parallel. Single stream -- and restore -- performance is significantly less impressive.
Backup systems with instant recovery allow your workloads to run using your backup data repository as their primary storage, or as a virtual machine right on the backup appliance. Either way, your backup repository has to be stored on disk rather than tape, and your backup application has to back up disk images, as opposed to the older file-by-file backup method. Since image backups are the norm for virtualization, and the basis for VMware’s vStorage API for data protection with changed block tracking, software vendors can add instant recovery by presenting the images they’re already storing.
My first exposure to instant recovery was when Veeam demonstrated its version, since dubbed vPower, at Tech Field Day 3 in 2010. For vSphere environments, the Veeam backup server presents the virtual machines being recovered to vSphere through an NFS server in the backup software. For Hyper-V, it loads an installable file system driver into the host. The latest version of Symantec’s NetBackup (7.6) has a similar instant restore feature for VMware VMs.
Instantly restoring physical machines is a bit more complicated. The newest version of DellAppAssure can push a driver to the machine being restored, which gives it access to the data in the backup repository as well as the data that’s already been restored.
Vendors that base their backup systems on appliances, such as Quorum and Unitrends, store backup images of both physical and virtual machines on the appliance and provide instant recovery using a hypervisor, frequently KVM, on the appliance. When a machine needs to be recovered, the appliance performs a P2V, or V2V, conversion on the system and spins it up on the appliance. This is basically the same technology that cloud disaster recovery services use to spin up a VM on a public cloud server.
The only real downside to instant recovery is that your backup repository is probably built using slow, high-capacity disk drives, so the recovered machine will run quite a bit slower than when using your shiny new hybrid array. From my perspective, having email or a mission-critical database up, but slow, in minutes is much better than waiting hours for data to be restored. Once the system is up and running, you can Storage vMotion or live migrate the workload to faster storage.
When chatting recently with the AppAssure folks about their new version, I had the brainstorm that using a server-side SSD as a cache would significantly ameliorate the performance loss that may be caused by running from the backup repository.
Disclaimer: Dell, Veeam and Symantec have been clients of DeepStorage, LLC in the past.