One of the challenges of virtualization is dealing with increasing number of server instances, especially in large data centers where they can easily get unaccounted for. Once the developers learn that it takes you an hour or two instead of days or weeks (including the purchase approval process) to provide a new server, the demand quickly increases. If you’re not careful and don’t set the provisioning rules in the beginning, one of them being that every virtual server, or at least every production virtual server needs to be backed up in accordance with your current policy and frequency as similar machines that have their own hardware box, backups can quickly grow out of control. The danger of managing backups by avoidance can be felt only when it becomes critical and someone needs a backup that doesn’t exist. It is true that some of this can be achieved with snapshots, but somebody has to double check whether backups exist and whether restoring works, otherwise the troubles will double when a restore is needed. Everybody likes the convenience and ease with which you can create new virtual machines. Nobody likes to back them up.
When we bought a whole new HP C7000 with a bunch of new blades and EVA4400 SAN to handle some of our virtual machines, we started looking into backup licenses and it turned out that the cost for backup agents was going to skyrocket. But then we did some research and through our vendor we got to talk to a group of people from another client of his who were already ahead of us on virtualizing their servers. Instead of buying individual backup agent for every single virtual server, they resorted to a much cheaper solution – they decided to use vmware scripts to pause, snapshot and back up all of the important VMs to a centralized depository on one server, then they back up that volume to tape using just single backup client license. Sweet.