Richard Jones at Data Center Strategies has brought up an interesting point. Linux does not have anything like the Volume Shadow Copy service that made it’s first appearance in Windows Server 2003 and has since appeared in WXP and everything since. VSS is a simple (the code probably is far from it) but powerful idea. It allows a file to be backed up while maintaining the integrity of the contents. This can then be used in a few ways. You can take regular backups and use a certain percentage of the volume for storing those backups, e.g. 10% of available space. Or your backup tool can use VSS as an open file agent to backup files to another media store, e.g. a dedicated backup disk drive or tape drive.
I started using VSS the week I first deployed W2003 back in the Summer of 2003. We enabled VSS on our file servers and never looked back. We allowed the file servers to use 10% of available disk space to keep as many backups as it could. We ran VSS twice a day, midday and in the evening. We also had file servers replicating (live) to DR site replicas and we backed up using Commvault every night (to disk and then to tape which was sent offsite). I think you could say we were protected 🙂
Why were we using VSS? I’m a believer in trying to do things quickly. Part of that is empowering users. For example, when procedures for file shares (who owns them, who has what permissions, etc, all on paper!) started slowing down responsiveness I started looking at SharePoint – we didn’t enable site creation delegation but we did allow admin delegation for the site owners. VSS gave us something similar. If a user wanted to recover a file, they simply used the Shadow Copy Client (for WXK and W2K, built into later OS’s). It integrates into Windows Explorer and backups are only a right-click away in the file share. Users could access any of the stored snapshots in Windows Explorer and using familiar windows, navigate to what they wanted from a time they wanted and copy it back to the file share. When a user didn’t know how to do it (not everyone is IT savvy and can be trained) then our helpdesk did it for them. The old experience of waiting hours for a backup to be catalogued and retrieved was gone.
Replication to DR helped in case we lost the site. We backed up to disk to speed up the backup process at night. We then streamed to tape to keep data for longer periods. Using 10% of available disk gave us anywhere from 1 week to 1 month of history depending on activity on the file servers.
MS extended VSS beyond just file servers. There are special helpers called VSS writers to assist with files that have complications. There are VSS writers for SQL, Exchange and Hyper-V. VSS backs up as much as the file as possible. There’s always just a little bit left over where changes where happening as the backup takes place. There’s an incremental process to capture these. The VSS writers allow a very quick stoppage of the service using the files so that the final bits of the backup can take place. The outage is so quick that no body notices. The integrity of the files is maintained with Microsoft’s support. Tools such as Data Protection Manager 2007 SP1, Iron Mountain LiveVault and Connected and Virtual Machine Manager 2008 VMM (for P2V) use this methodology.
The blog post points out that this is something missing from Linux. I’d have to agree. I actually find it surprising that they haven’t come up with something similar. VSS is a very powerful and under-marketed engine.