Paragon Software Group

Konstantin Komarov

We get lots of questions about the necessity of backups with respect to virtual systems such as VMware virtual environments or ESXi hypervisors. The great number of existing technical terms on the topic and the equally great number of solution approaches make it hard to get an overview of it all. While snapshots and “classic” backups are related in nature, there are fundamental differences. And then there is also replication as a further member of the big family of data protection, backup, business continuity, and high availability.

Hardware virtualization offers many advantages when it comes to making the most cost-effective use of expensive IT infrastructure. Furthermore, virtual systems are ideally suited for maintaining high availability of critical business applications and data. High availability and data backup are also closely related, but they are by no means identical twins.

Special case: Agent-based Protection of VMware ESXi Guests

vSphere Backup for Long-term Protection

Replication and Backup of vSphere Guests

Automatic Replication of vSphere Guests

Which is Better? Backup, Snapshot or Replication:

So when it comes to the question—whether backup or snapshot is the better choice, the simple answer would be: “Neither of them!” It simply depends on the scenario at hand and the platforms being used, so it’s clear that the answer’s not so simple after all. Snapshots are ideal for making quick changes to the virtual machine. These changes can then either be used or discarded with just a few clicks. In the latter case, the snapshot is “played back” on the ESXi machine. Strictly put a snapshot is thus neither suitable for high availability nor backup scenarios—its importance lies in the field of maintenance. In order to keep a virtual machine highly available, it’s a good idea to create VM replications, because the replica of a virtual machine corresponds to the original and can take over its task in just a few seconds. Replicas and snapshots are both temporary—so they are not suitable for retrieving lost datasets from the distant past.