In nonvirtualized environments, backups are usually done by installing a software agent in the operating system that performs a file-level copy of the entire environment. Examples of this approach include Symantec NetBackup and Backup Exec.
But in highly consolidated virtual environments, inserting agents and coordinating backup schedules can be tricky, and the agents can drain the CPU and I/O resources slated for other virtual machines (VMs).
These days, the preferred method is to create a snapshot of a Virtual Machine Disk File (or VMDK in VMware parlance, and the Virtual Hard Disk for Microsoft Hyper-V and XenServer environments), said Jerome Wendt, a lead analyst at DCIG in Omaha, Neb. By working off a snapshot, the backup server relieves the production server of the data-processing burden, minimizing the impact of the backup job on other VMs on a host.Snapshots step up VM backup process
There are variations on the snapshot-based backup theme. VMware's own VMware Consolidated Backup, for instance, quiesces virtual machines on an ESX host and makes them available to third-party backup software providers for off-host processing. Another approach is to use the snapshot capabilities of a storage array.
That's what UniSource Energy Corp. subsidiary Tucson Electric Power (TEP) does. TEP uses the snapshot family of products on its NetApp storage arrays to back up 300 virtualized Windows machines running on about 50 VMware ESX hosts, said Stephen Kiser, a systems administrator at TEP.
Specifically, TEP uses NetApp's SnapShot, SnapVault and SnapMirror to perform local backups and archives and to replicate them off-site. "We keep a couple of snapshots local for a couple of days for speed [of restore]," he explained. "Anything older than that we ship out to a DR [disaster recovery] facility," he added.
The decision to use snapshots was borne of necessity. With NetBackup, "our backup window was exceeding 24 hours," Kiser said, meaning that TEP could not complete a daily backup of all its systems.
Moving to a snapshot approach also cost TEP a lot less, said Chris Rima, a TEP supervisor of infrastructure systems. Even though NetApp charges for its SnapShot options, it is definitely cheaper than NetBackup for 300 systems, Rima said.
TEP doesn't bother to quiesce its VMs before taking a snapshot, arguing that the likelihood of the snapshot not being able to produce a crash-consistent image is exceedingly slim. And even if the first snapshot fails, they go back to the previous one. "Your chances are way above 99% that your application is going to come back up," Rima said. "We recover from [snapshots] all the time," he said. "We know it works."Improving speed, reducing costs with better backup
Another VMware shop that has thrown away its old-school backup software is Budd Van Lines, a Somerset, N.J.-based moving and relocation firm. In 2006, backups performed using the agent-based CA ARCserve took all night to complete, said Doug Soltesz, the vice president of information systems and technology. "When you're completely virtualized, you don't want agents in the VMs," he said.
Soltesz tried out Vizioncore vRanger and PHD Software's esXpress and found the latter to be faster and less expensive. Like other virtualization-centric backup suites, PHD esXpress backs up the entire virtual machine, but it does so using a small virtual machine running on the same host as the virtual machines it backs up. PHD esXpress spawns as many of these so-called virtual backup appliances (VBAs) as it needs to meet its backup window and uses throttling to prevent them from overconsuming resources.
Compared with ARCserve, esXpress completed its backups by 2 a.m. instead of the early morning, Soltesz reported. Also, since installing the latest 3.5 version, backups complete even faster -- by 12 a.m. -- thanks to new delta technology that sends over only recent changes. New data deduplication and file-level restore capabilities in version 3.5 allow the company to keep more data on disk for longer, and to recover individual files from a backup rather than having to restore an entire VM, Soltesz said.
Traditional backup software providers, meanwhile, have responded to the rise of virtualization largely by integrating with VCB and backing up the entire VMDK, said DCIG's Wendt. They plan their backup jobs by querying all the ESX hosts about the virtual machines running on them. However, in large virtualized environments, "that can get cumbersome," he said. The next frontier for backup software is to communicate directly with vCenter, VMware's management console.