Backup agent
A backup agent is software installed on a protected server or virtual machine that reads its files, application data or changed disk blocks and sends them to backup storage. Running inside the machine lets it put applications into a consistent state before the copy, which a read from outside the machine cannot always do.
What does an agent do that an outside read cannot?
An agent has three abilities that come from running on the machine itself. It can ask a database or application to flush pending writes and hold a consistent state, so the backup restores to a working database rather than a file that needs repair. It can track which files or disk blocks changed since the last run, on some operating systems through a small driver in the kernel that sees each write, so each incremental backup copies only the difference. And it can compress and encrypt data before it leaves the machine.
Agents take their instructions from the backup software server, which schedules each job and records the result in the backup catalog.
How does agentless backup work?
Agentless backup reads a virtual machine through the hypervisor, using the hypervisor's snapshot and changed-block tracking features, with nothing installed in the guest. Application consistency then depends on guest tools the hypervisor already uses, or on a helper process the backup software runs inside the guest only for the length of the job. Physical servers have no hypervisor to read from, so they are outside its reach.
Agent and agentless trade-offs
| Property | Agent-based | Agentless |
|---|---|---|
| Where the backup code runs | Inside each protected machine | On the hypervisor or a backup proxy |
| Physical servers | Supported | Not supported |
| Application consistency | Direct, through the application's own interfaces | Indirect, through hypervisor snapshots and guest tools |
| Maintenance | One install per machine to patch and keep in step with the backup server | Nothing to maintain inside guests |
| Load during the job | CPU, memory and disk reads on the production host | Shifted to the hypervisor and its snapshots |
| Attack surface | A privileged service on every protected machine | Concentrated on the hypervisor and backup server |
Many products combine the two: agentless for virtual machines, agents for physical servers, databases and some cloud instances.
Why are agent hosts a credential risk?
An agent runs with high privilege on production machines, since it has to read everything on them. If the host also holds keys that can write to and delete from the backup repository, an attacker who controls the host can remove the backups it created. Where the agent hands data to a backup server or gateway that holds the storage credentials, a compromised host can submit new data but cannot remove old restore points. Keys left in agent configuration files are one of the routes from credential theft to lost backups.
Frequently asked questions
Do all backup products use agents?
No. Many are agentless for virtual machines and cloud workloads and use agents only for physical servers, databases and applications that need in-guest coordination.
Can an agent make a backup without stopping the application?
Usually yes. It asks the operating system or application for a consistent snapshot or log checkpoint, reads from that and releases it, so the application keeps running with a brief pause in writes.
Does the agent need the backup repository password?
Not in most designs. The agent needs enough access to hand data to the backup system; where it also holds delete rights on the repository, the backups are only as safe as that host.
