"I don't think we're seeing enough discussion around image sprawl," said Andi Mann, Enterprise Management Associates VP of research, storage management. "We're starting to see snapshots and image sprawl as a major storage issue. The idea that you can restore critical data, from a compliance perspective, it's overkill. You end up storing and restoring 10Gb even though you want to get at one file."
So how many snapshots should a storage manager save? "There's not any one rule for that," said Lauren Whitehouse, Enterprise Strategy Group analyst. "It's so environment-specific, use-case-specific, and it could even be compliance-specific."
Whitehouse said that snapshots are just another technology adding to an overall storage capacity problem. "This is a problem that goes beyond just snapshot copies," she said. "There are too many copies of data made and then data is also held on to for longer than it's worth. The more I talk to users, the more I find there's a certain amount of apathy around the whole lifecycle management of data. Unless they are mandated to destroy it, they won't do it."
Tim Doyle, network administrator and IT manager at Pittsfield, MA-based Legacy Banks, manages his snapshot data using volume limitations built into Windows Server 2003's Volume Shadow Copy Service. Legacy Banks currently saves 41 MB worth of snapshots, which is one percent of the total volume size. "Most of it's fire and forget," said Doyle. "We'd probably only revise it if we realized it was eating up too much disk space, or if we tried to get something back and we were unsuccessful, if we discovered it was copying over the last day's snapshots."
"I feel 100 percent confident that this is not compromising the system," he said. Doyle uses differential snapshots for near-term data recovery. "I would prefer to see at least a week's worth of snapshots," he said. "That's a personal preference; it's not mandated." He sends tapes off site daily, so snapshots recover data for users quickly. "If there's a versioning problem with the only copy, it's not acceptable to tell someone to wait 24 hours to get the tape back," he said.
For now the free product meets the company's needs, Doyle said. "SAN software for snapshotting gives you flexibility for doing offsite cloning for disaster recovery and volume recoverability because you're moving it off volumes," he said. But, "you're talking about a very expensive proposition to do that." Doyle isn't sure he'd be able to recreate a volume from snapshots in a disaster, since the shadow copies would likely be gone.
Snapshots as an aid in disaster recovery
Mann said that snapshotting can sometimes do the work of rebuilding after a disaster. "Even if it's only a differential, you can apply multiple snapshots until it's the full machine, the latest copy," he said. "But at some point it's going to be faster to rebuild rather than apply incremental snapshots."
"The efficient forms of snapshots are differentials," said Eric Schott, director of product management for Dell EqualLogic, which bundles its Auto-Snapshot Manager software into its arrays. He likes to call the method of creating snapshots in the array "allocate on write" rather than "copy on write." According to Schott, "by default, automated methods of creating snapshots have automated methods of cleaning up snapshots." But customers still need to plan carefully to decide how much space to allocate for snapshots.
"Sometimes they have to think about that a little more," he said. Dell's setup means that snapshots reside in a separate reserve that's configured for each volume, and users set policies to allocate the percentage of a particular volume that stores snapshots -- with an upper limit of 512 snapshots.
Vinu Jacobs, who handles the storage infrastructure at a New York City-based law firm, manages his snapshot data with a strict retention policy that's based on extensive data modeling and planning. "It took a little bit of work on our part" to decide how much snapshot data to save, he said. "We did the legwork up front so that the snapshot pools were the right size."
"We have done rebuilds from our snapshots but they've generally been more on a test basis," said Jacobs. "From a production perspective it's been much more related to a data recovery. We as a firm went into this with the mindset that we wanted, even if we spent a few extra dollars, disaster avoidance rather than disaster recovery."
Mann thinks that snapshots have their place, but aren't the cure-all for future storage needs. He says that snapshots aren't -- or shouldn't be -- data backup. "Snapshots are good fast technical way to take a copy of a machine. That's not data backup and that's not data archiving."
Mann sees snapshot management as intricately related to the challenges of managing virtualized servers -- challenges he says most organizations are still struggling with. "Image management is lagging behind the deployment of virtualization technology," he said. VMware's Consolidated Backup (VCB), for example, doesn't clean up snapshots automatically, but rather requires third-party tools for storage management. "We need to use backup solutions that understand the nuances of virtualization," said Mann. "Snapshots are the easy road and unfortunately that's where we're going. There's room for snapshots to get a little smarter, but they're not the solution."Christine Cignoli is a Boston-based technology writer. Visit her at www.christinecignoli.com.
Dig deeper on Remote data protection