Agent or Agentless Backups of Virtual Machines

Over a decade ago, virtualization was just starting to take hold and using an agent to perform backup operations on a client was an unquestioned standard. Today, technology has given us a choice to perform agent or agentless backups of virtual machines.  But which is better?  The question is simple but the answer can be complicated.  Let’s take a look at some suggestions and examples to help backup administrators decipher what’s best for them.

First and foremost, to consider agentless backups of virtual machines, the backup product in use must support integration with a hypervisor. The top three type-1 hypervisors available are VMware, Hyper-V and Citrix XenServer, so we’ll focus on those.  Typically, the agentless backup process will leverage the hypervisor snapshots, and possibly a change block tracking mechanism to perform incremental backups.  Unitrends supports agent or agentless backups of all three major hypervisors.

At a high level, the more servers there are to protect, the more agentless backups provide an advantage, simply because there is less to administer and fewer agents to install or update. In addition, agentless backups can have less of an impact on system resources, whereas an agent could consume valuable CPU, memory or storage I/O, if resources are tight.  However, agents have their advantages too, and sometimes not all VMs can be protected via hypervisor snapshots.  Since the agentless process involves snapshots, the VM must support snapshots and include any data disks that need to be protected.  Therefore, if a VM includes physical raw disk or storage that is not included in a snapshot, a traditional agent based backup would be the best approach.  We also shouldn’t overlook the technical details, such as how content aware deduplication relates to backup types.   In this case, agent (File) level backups will provide the most efficient deduplication rates.  You can read more about Unitrends content aware deduplication on a related blog post, here.

Digging a bit deeper, agentless backups typically have less visibility into the server than an agent would, and therefore could be considered less granular. Any modern backup and recovery product with agentless backups will support file level recovery from either backup method.  However, file and folder level indexing or file level exclusions and inclusions are typically agent specific.  This makes a large file server a possible candidate for an agent based backup.

Server and application vendors may also play a role in this decision. Some applications may need direct agent interaction to complete an application consistent backup or to perform log truncation.  An example of this might be SQL databases with transaction logging.  In order to backup the database and properly truncate SQL transaction logs, an agent based backup would be the best method.  In addition, the ability to separate application level backups from the server backup can be very important in meeting business RPOs.  For example, an agent could allow the server to be backed up once a day while the critical business application is backed up every 5 minutes.  No excess data is backed up, so backup windows are short and RPOs can remain as small as possible.

Domain controllers are also another often overlooked candidate for agent based backups. Even though we can successfully protect a domain controller with either backup approach, Microsoft dictates that every domain controller has an agent (In-Guest) based backup that includes a backup of the Active Directory database.  (TechNet Reference: Here)  This makes an agent based backup of Windows Domain Controllers the only “supported” option.

From a recovery standpoint, product options will vary between backup and recovery solutions. With Unitrends, we offer full system level recovery with either agent or agentless backup approaches.  In addition, granular file level and even instant recovery options are supported with either approach.  Typically the agentless backup will provide the fastest recovery times for full system recoveries and an agent based approach will allow for the greatest range of compatibility.

In the end, the most flexible backup and recovery solutions provide the ability to leverage both techniques. At Unitrends, we refer to this ability as “Deep Virtualization,” and discuss it in detail, here.  Generally, a mix of both agent and agentless backups provide the best overall protection and recovery options, while providing complete coverage in heterogeneous environments.  Ultimately, the choice will be yours.

If you have any other good examples and would like to share your thoughts, we’d love to hear from you.