Windows Server Backup is the native block-level backup tool, and it writes through Volume Shadow Copy Service. That single design decision explains nearly every operational characteristic of the product: backups capture a point-in-time image rather than an open file stream, VSS needs free space on the volume it snapshots, and differential backups grow until an intervening full backup resets the chain. Administrators who treat it as a file-copy tool end up with failed jobs, and administrators who treat the differentiation logic as opaque end up with restore times they did not plan for.
This guide covers a working rotation that fits the tool’s actual behaviour: a weekly full, daily differentials, a dedicated backup target, a shadow copy space allocation sized for the number of retained snapshots, and a restore drill you can run in under an hour. It assumes the Windows Server Backup feature is installed and that you have a second volume or share available as the target — our Windows hosting comparison covers what storage capacity to look for if you are sizing the host.
How VSS Determines What a Backup Costs
A VSS snapshot is copy-on-write: when a block on the source volume is modified while the snapshot exists, the original block is copied to the shadow storage area before the write completes. The cost of a snapshot is therefore proportional to the write churn during the backup window, not to the size of the data being backed up. A 200 GB volume that changes 2 GB during a backup needs roughly 2 GB of shadow storage, plus headroom for writes that continue while the snapshot is being read.
Shadow storage is configured per source volume with vssadmin resize shadowstorage, and the default on a new volume is often unset or small. If the shadow storage fills during a backup, VSS deletes the oldest shadow copy — which can break a backup mid-stream — or the write itself fails with a 3033 or 3034 error. Check the current allocation with vssadmin list shadowstorage before you schedule anything, and set it to at least the size of your expected churn times three.
Full vs Differential: What the Chain Actually Does
A full backup of a volume copies every used block and resets the differential chain. A differential backup copies every block that changed since the last full backup, which means each successive differential is larger than the previous one. A weekly full on Sunday and daily differentials means Monday’s differential is small and Saturday’s is nearly as large as a full. This is the defining property to plan around: capacity and backup duration both grow through the week.
Windows Server Backup schedules this through the wbadmin command line far more reliably than through the GUI. Run wbadmin start backup -backupTarget:E: -include:C: -allCritical -quiet for the weekly full and wbadmin start backup -backupTarget:E: -include:C: -allCritical -differential -quiet for the daily jobs, each driven by its own scheduled task. Do not let the GUI’s single-schedule dialog decide for you; it cannot express a full-plus-differential cadence without a scheduled task behind it anyway.
Sizing the Backup Target and Retaining Generations
The target needs to hold the current full, the largest differential in the chain, and at least one previous generation if you want a rollback point. With a 100 GB used footprint, that is roughly 100 GB plus 60 GB plus 100 GB — about 260 GB, before compression. Windows Server Backup compresses on the target by default for volume backups, often by 40 to 60 percent on typical server workloads, so a 200 GB target is a reasonable starting point for this example and a terrible one if the data is already compressed media files.
Retention is not automatic. Windows Server Backup will overwrite the oldest backup when the target fills, which means a single misfiring job can silently consume your history. Either monitor free space on the target with a scheduled check that alerts below twenty percent, or script a rotation that removes generations older than a defined age with wbadmin delete backup -keepVersions:N. The latter is explicit and auditable; the former is a safety net.
Application Consistency and -allCritical
VSS coordinates with writers: SQL Server, Exchange, Active Directory, and IIS each register a writer that quiesces its data before the snapshot and resumes after. This is what makes a VSS backup restorable rather than merely consistent with the file system. Including -allCritical covers the volumes that hold operating system state and the boot configuration, which is what makes a bare-metal recovery possible rather than a data-only restore.
Check writer health before every scheduled run, because a failed writer silently downgrades the backup to crash-consistent for that application. vssadmin list writers should show every writer in the Stable state with no errors; a writer stuck in Waiting for completion means the previous snapshot did not complete cleanly, and restarting the corresponding service is usually the fix. A backup job that reports success while a writer was failed is the most dangerous outcome in this entire workflow.
The Restore Drill
A backup is a hypothesis until you restore from it. Run the drill quarterly against a non-production instance, and run it immediately after any change to the schedule, target, or included volumes. Mount the backup as a virtual disk with wbadmin start recovery -version:<timestamp> -itemType:Volume -items:C: -recoveryTarget:E:\restore, or recover individual files by their full path. The point is to measure elapsed time, not to prove that the files exist.
Record three numbers from each drill: time to mount or recover the volume, time to boot and validate the application on the recovered instance, and the gap between the backup timestamp and the last application transaction visible after recovery. If the recovery point objective your application team believes they have is four hours and the drill shows a twelve-hour gap because the differential chain was misconfigured, that is a finding worth surfacing before an incident rather than during one.
Monitoring and Common Failure Signatures
Windows Server Backup writes to the Application log with source Backup. Monitor for event 4 (backup started), 14 (backup completed), and 517 (backup failed) and alert on the absence of 14 within the expected window as well as on the explicit failure. A missing success event is often the first signal that a scheduled task stopped running because a service account password changed.
The recurring failures follow a short list: 0x8078002A means insufficient disk space on the target; 0x807800C5 means the shadow copy could not be created, usually a writer problem or exhausted shadow storage; 0x800423F4 indicates the volume is not supported as a backup source. Map each to a first action in your runbook and you will resolve most failures without escalating. If you are rebuilding the host rather than tuning an existing one, the provisioning walkthrough on our main page covers separating the backup target onto independent storage from the outset.



