But which one is best suited to a particular environment? In order to answer this question, we will have to explore the details of the two approaches. For the sake of simplicity we will assume that our goal is to perform file-level backups for Windows virtual machines.
What are snapshots?
Both VCB and SAN-based backups are based on snapshots. Once the snapshot is taken, it can be backed up while the application continues to run on the original volume. The backup window constraint becomes irrelevant.
The virtual machine snapshot is a picture of the virtual machine at the time the snapshot is taken. Besides the image of the virtual hard disk, the snapshot includes the virtual machine configuration file, BIOS configuration, RAM content and the content of various processors.
Once the snapshot is taken, the image of the virtual disk can be mounted in read-only mode to a local directory on the VCB proxy server. The files present on the virtual disk can be archived by a backup agent installed in the VCB proxy server.
Storage snapshots are different than the virtual machine snapshots. It is a point-in-time copy or exact duplicate of a data volume taken at a specific moment. A storage snapshot is highly efficient as no data copy is actually performed; the data that remains unchanged is only stored in one place and is referenced both by the baseline volume and the point-in-time copy. A storage snapshot is can be accessed from a virtual machine via a dedicated LUN, provided by the storage array. A backup agent installed in the virtual machine can archive the files.
Choosing the right snapshot option
Virtual machine data can be stored either in a virtual disk file or directly in a SAN LUN. Both storage options are compatible for physical and virtual modes. If the virtual machine uses virtual disk files, you need to ensure that your SAN supports snapshots for the file system in place (VMFS or NAS). If not, then you must use VCB snapshots. Please note that all virtual disk files present within a volume will be part of the snapshot. For this reason, it is recommended to regroup virtual machine disk files with similar backup policies in one volume.
If the virtual machine uses raw data mapping (RDM) in physical mode, then your only choice are SAN-based backups as virtual machine snapshots are not possible. If the virtual machine uses raw data mapping in virtual compatibility mode, then both choices are available. Virtual compatibility allows the LUN to behave like as a virtual disk file. Besides the LUN volume, you need to back up the RDM mapping file.
Remember, VMware snapshots are software-based. As such, they have scalability and performance issues. The VMware Consolidated Backup (VCB) uses a dedicated physical server to connect to the available snapshots and back up their contents to tape. This will have an impact on the production disk subsystems. Storage snapshots have little impact on performance and are nearly instantaneous. The granularity of the stored changes is at the block-level, resulting in a very space-efficient method.
Keeping data consistent
With snapshots, the data is captured in a disordered state. Application recoveries from this state may not be successful. For this reason, both VMware snapshots and storage snapshots require the virtual disk(s) to be in a restore-consistent state.
This is achieved by making both the application and the underlying operating system aware of the snapshot process. An application-specific script prepares the application for the backup by flushing all buffers and committing all data. Next, a "sync" command issued by VMware tools flushes the operating system buffers to disk.
Both VCB and SAN-based backups provide an interesting alternative to traditional backups. Besides the technical arguments, the backup approach may be ultimately decided by the choice of the team responsible with corporate data availability.
This was first published in April 2008