Australian cloud for business software — built and run in Sydney since 2002

NetLeverage
NetLeverage

Image-level vs file-level backup: what each restores and what it leaves behind

Technical article
Backup & Recovery

About this article: This article is reviewed against the cited sources. This is source research, not lab validation. Check the stated versions and your own requirements before making changes.

An image-level backup copies whole disks or volumes, including the operating system, the disk layout and the files, so it can bring a server back as it was. A file-level backup copies the files and folders you chose, so it can bring those back and nothing else. Each leaves something behind. An image can be heavy to use for one lost document. A file backup cannot rebuild a server.

If a backup product is sold by one of these labels, the useful question is not which is better. It is what you need on the worst day: one file, one folder, or the whole server.

What does an image-level backup capture?

The unit is the disk or volume. On a physical Windows server, Microsoft's wbadmin start backup shows the idea: with -allCritical it includes "all critical volumes (volumes that contain operating system's state)", described as "useful if you're creating a backup for bare metal recovery". With -systemState it adds boot files, the registry, SYSVOL, and on a domain controller the Active Directory database.

On a virtual machine the unit is the virtual disk plus the machine's definition. Microsoft's Hyper-V documentation separates two kinds of backup. A "VM backup" backs up "the entire virtual machine, including the configuration and all data" and each full or incremental backup "must copy over all associated virtual machine configuration and virtual machine state files". A "data-only backup" backs up only what is stored on the virtual hard disks, and to restore it "you create a new virtual machine and restore the data to it."

On VMware, Broadcom's guidance for backup software describes the sequence: take a snapshot through the vSphere API with the quiesce flag so the guest is "a static (quiesced) view" while it keeps running, "capture the virtual disk data and virtual machine configuration information", read the disk and snapshot files, then delete the snapshot.

What does a file-level backup capture?

The unit is a path. You choose folders, the tool copies the files under them, and a restore puts files back. wbadmin start backup -include:<items> accepts "multiple files, folders, or volumes", so Windows Server Backup can do either; the choice you make decides what you hold.

The file backup holds file data and, if the tool supports it, permissions. For wbadmin start recovery, Microsoft documents that "by default, the security ACLs are restored" and that -notRestoreAcl skips them. Other products vary, so ask.

What does each restore, and what does it leave behind?

Image-level File-level
Restores a whole server or VM Yes, if the image includes the system volumes and boot state No: it holds files, not the disk layout, boot files or installed software
Restores one deleted file Possible, if the tool can read files inside the image; some restore the whole volume first Yes, the normal use
Holds the operating system and installed programs Yes (system volumes) No, unless you include system folders, which is not the same as a bootable system
Holds settings stored outside the files you chose (registry, shares, services) System state or whole volume brings them Only if you back them up separately
Cost More data read and stored Less, because only chosen paths are read
Main limit Hardware or hypervisor differences at restore; large restores Everything not listed is not there

Windows Server Backup shows how the same tool crosses the line. wbadmin start recovery -itemtype:File restores files and folders from a backup; -itemtype:Volume restores a whole volume; -itemtype:App restores an application registered with the tool. Microsoft says wbadmin must be run from an elevated command prompt, and the system recovery command wbadmin start sysrecovery "must be run from the Windows Recovery Console". That second command is the whole-server route: it restores "critical volumes" by default, and -restoreAllVolumes is needed to bring back the others.

What does an image restore leave to you?

A restored image is a copy of the machine as it was when the backup ran. That includes its mistakes. Three things to check:

  • Hardware differences. A physical server restored to different hardware may not boot or may need drivers. "Bare-metal recovery for Windows Server: VSS, system state and dissimilar hardware" covers this.
  • Application state. Whether an application is consistent in the image depends on how the snapshot was taken. The VMware guidance quoted above uses a quiesced snapshot; Microsoft says Hyper-V WMI-based backup "still utilizes VSS inside the virtual machine". For databases, that is a test to run, not a label to trust. The guide to application-consistent VMware backups explains how.
  • Everything newer than the backup. Restoring a whole volume replaces what is there. wbadmin start sysrecovery with -recreateDisks warns that it "deletes all data on volumes that host operating system components" and "might also delete data from data volumes".

Risk: a whole-volume or -recreateDisks restore overwrites the current disk contents. Microsoft states that -recreateDisks deletes data. There is no way back for what was on those volumes unless they were backed up separately first; use -excludeDisks (valid only with -recreateDisks) to leave named disks untouched.

What does a file restore leave to you?

To get a working file server back from file-level backups alone, you first need a server: an installed operating system, drivers, the file-serving role, domain membership, network settings and the share definitions. Then you restore the files. The guide to restoring a file server covers shares and settings. A file backup that has never been tried against an empty machine has not shown it can do this.

Example (illustrative): fs01 is a virtual machine with an 800 GiB disk, of which 300 GiB is used. The shared data is 250 GiB under D:\Shares.

  • With an image-level backup of the VM, a total loss is restored by recovering the VM from the backup, with the operating system, shares and configuration inside it. The time is dominated by moving the stored data. At an assumed 100 MiB/s, 300 GiB is 300 × 1,024 = 307,200 MiB, or 3,072 seconds, about 51 minutes, if the tool restores only used space.
  • With a file-level backup of D:\Shares, the same 250 GiB at 100 MiB/s takes 250 × 1,024 ÷ 100 = 2,560 seconds, about 43 minutes. That figure covers only the data copy. The server itself must first be built and configured by hand.
  • A user who deletes one spreadsheet is better served by the file-level backup, or by an image backup whose tool can browse files inside it.

How can you check which kind you have?

  1. Read the job definition. Does it list volumes or disks, or folders? On Windows Server Backup, wbadmin get versions lists available backups; Microsoft documents it as the way to find a version identifier. wbadmin get items is documented to list items in a backup. Check both read-only.
  2. Ask what a restore of the whole server needs. Does the provider restore a virtual machine, or only files into a machine you rebuilt?
  3. Ask if single files can be restored from the image without restoring the whole disk first, and how long that takes.
  4. Test on a scratch machine. Restore into an isolated virtual machine, not the live server. Boot it, open an application, and read back a known file. Do the same for one single-file restore.

Which should you choose?

A plan benefits from both abilities, whether from one product or two. Image-level protects against loss of a server; file-level or file-from-image restore handles the daily "I deleted a folder". What matters is that you know which ability you have actually tried.

A short way to decide is to write down three recovery cases and match each to a method:

  1. A person deletes or overwrites a document. You need a recent restore point and an easy way to pick single files. File-level backup, or an image backup whose tool can browse inside the image, covers this.
  2. A server will not start, or its disk is lost. You need the operating system, the disk layout and the data together. Only an image-level backup that includes the system volumes covers this without a rebuild.
  3. A site or storage system is lost. You need a copy in another location, in whichever form, plus a place to restore it to. The method matters less than where the copy lives and whether anyone has restored from it.

If a provider can show you a restore for each of the three cases, the label on the product is secondary. If it can show you only one, you know which gap to ask about.

Risks and limits

  • These commands have not been run or lab-tested. They are taken from Microsoft's wbadmin documentation and Broadcom's backup guidance.
  • Microsoft's pages show 2023 and 2025 dates; the Broadcom page shows none. Other vendors' products can capture and restore differently; the vendor's documentation is the source for its product.
  • The existence of wbadmin get versions and wbadmin get items is taken from the wbadmin pages' own references; the output formats are not described there.
  • Whether a particular tool can restore a single file from an image without restoring the volume varies and the documentation does not state it.
  • The timings are arithmetic on assumed 100 MiB/s speeds; real restores depend on the network, storage and tool.

NetLeverage includes a daily full disaster-recovery backup of each hosted server at no extra charge. Talk to an engineer.

Sources