Virtual servers can be a very effective way of getting the most bang for your hardware buck. Because you can run multiple virtual-machine (VM) instances at the same time on shared hardware — each behaving as a completely independent OS — server resources need never be idle. On a large scale, this can mean a significantly lower investment in physical iron, not to mention associated costs like power and cooling. But packing four, eight, or even 20 or more virtual machines on a single server means that there are a lot of eggs in one basket — and those eggs are your data.
A virtual machine image comprises two basic kinds of data — the operating system itself, including its core files, update patches, and files it generates like logs; and application or user data, or everything related to the apps that run on the virtual machine. You can even break down the latter into the installed apps and data they generate — for example, a mail server app and the email messages themselves.
In practice, your organization must begin with an inventory of which kinds of data are critical to backup. Remember that running frequent, scheduled backups can generate a lot of data — both on the network and in storage resources. Do you need to backup the entire OS, the installed apps, and the data? The answers will vary widely depending on your situation and the software being used.
There are three basic recipes for backing up virtual servers. Each has its own distinct advantages and drawbacks, so the key best practice is for your organization to understand its backup needs and choose which approach is best suited to meet those needs.
Guest-based backups: Each virtual machine instance is a guest. Because the guest itself is a full-fledged operating system, you can run traditional backup tools within that OS, just the same as if the OS were running on dedicated physical hardware. One plus to this approach is that if your organization already has a preferred backup tool and has developed strategies and scripts for it, there is no extra learning curve. Another advantage is that the backups generated this way will be very granular, meaning that you can easily recover individual files should the need arise. On the con side, if your backup tool requires a license for each installation, costs could rise dramatically, depending how many virtual machines you run. The backup data itself needs to be stored somewhere, possibly outside the virtual machine's own storage. This could generate a lot of network traffic. Lastly, running frequent scheduled backups within the guest VM could increase the load on the physical machine, somewhat undermining the value of stacking VMs on shared hardware.
Host-based backups: From the point of view of the physical host OS, each VM is a collection of files. There is a configuration file that defines the VM environment, and usually a series of data files which emulate physical storage. Therefore, the files that make up a particular VM can themselves be backed up from the primary OS. Only one license would be needed to use the backup tool, because it is running only one instance, even though it can "see" all the installed VMs. Because the entire VM is being backed up, restoring a VM in the case of disaster recovery is a very quick process. However, it may be difficult or impossible to restore individual files from the VM's virtual file system. To preserve data integrity, the VM would need to be in a shutdown state before the backup process begins. This may mean custom scripting to shutdown and restart the VM, not to mention downtime for whatever services that VM provides. On a host with many VMs, it may consume a lot of resources to generate regular backups of all of them, undermining VM performance itself. In some cases, high-performance virtual machine connect directly to a physical storage system rather than generate a virtual file system—in this case, some backup tools may not be able to backup the entire VM at all.
Proxy-based backups: Simply put, this solution involves a dedicated backup server. The server is a physical machine (not a VM) with software that connects to your physical hosts and controls the virtualization software. The dedicated server can then take snapshots of live VM's and backup all or specified portions of the data within, saving the data to built-in or network storage. Products like VMWare vSphere and Microsoft's Hyper-V backup server can centrally manage backups from a large cluster of VM hosts, without the inherent limitations of either guest- or host-based backups. But a proxy-based backup system is also the most costly, both in resources (such as the dedicated server), setup, and maintenance.
For more information, see:
Protecting Users With Virtualization
SSD in the Enterprise: The New Face of Tiered Storage
Five Disaster Recovery Technology Trends
        | 
 |
