A common calculation when buying a backup repository: the source data is 10 TB, “let’s get three times that,” 30 TB. A year later, as monthly GFS points pile up, the repository fills and nightly jobs start failing with insufficient free space errors. The real need differs many times over depending on the backup chain type, daily change rate, retention policy, data reduction ratio and the file system’s block cloning support. Below we go step by step through these variables, a worked example table and the method for monitoring capacity afterwards.
The variables that determine capacity
The size of a backup repository is calculated with five inputs. Skip one and the result is wrong.
- Source data size: Not the allocated size of the VM disks, but the amount of used blocks. In environments with thin disks, the difference can be substantial.
- Daily change rate: The amount of unique blocks that change within a backup interval. It directly determines the size of incremental backups.
- Data reduction ratio: The reduction from compression and in-job deduplication together. It varies a lot by data type; already compressed media, encrypted files and database backup files hardly shrink at all.
- Retention policy: The number of short-term recovery points and the number of GFS (weekly, monthly, yearly) full backups.
- Backup chain and file system: Determine whether synthetic full backups physically take up space.
On top of that, add growth headroom and working space (free space temporarily needed during merge operations). A repository that forgets working space fills up during a merge.
Measuring the change rate
The input that moves the calculation the most is the change rate. Don’t guess; measure it in the existing server and system infrastructure. Practical methods:
- History of existing backup jobs: The most reliable source. Note the “transferred” sizes of incremental backups over the last two to four weeks, with the values before and after data reduction. Look separately at peak days such as weekends and month-end close; you need space to cover the peak, not the average.
- Hypervisor changed block tracking: VMware CBT and Hyper-V RCT track changed blocks; the backup software calculates the incremental size with this information. If there are no backups yet, running a short trial job gives more accurate results than watching performance counters for weeks.
- Reporting tools: Monitoring products such as Veeam ONE include VM change rate estimation reports.
Disk write counters (for example LogicalDisk\Disk Write Bytes/sec on Windows) overstate the change rate. Even if the same block is written a hundred times during the day, it takes up space only once in the incremental backup. These counters are only an upper limit.
Backup chain types and their effect on storage
The same retention policy takes up different amounts of space with different chain types and puts different I/O load on the repository. You can’t calculate capacity without choosing the chain type.
| Chain type | How it works | Storage impact | Performance impact |
|---|---|---|---|
| Forever forward incremental | One full + incrementals; the oldest incremental is merged into the full every day | Lowest | Daily merge read+write load |
| Forward incremental + synthetic full | On certain days a new full is built from the data in the repository | Without block cloning, each full takes extra space | Source unaffected, repository loaded |
| Forward incremental + active full | On certain days all data is read again from the source | Each full takes physical space | Source and network loaded; the chain becomes independent |
| Reverse incremental | The newest point is always a full; old blocks are moved to rollback files | Low | One read and two writes for each changed block; the method that strains the repository most |
A synthetic full backup builds a new full file from the existing full and incremental files in the repository. On a classic file system (NTFS, ext4/XFS without reflink) this means physically copying the data: it takes hours, and each synthetic full takes up the space of a full backup. An active full backup, by contrast, restarts the chain from the source. It produces an independent copy against silent data corruption, at the cost of straining the source system and the network. It isn’t free.
GFS retention policy
The GFS (Grandfather-Father-Son) approach adds increasingly sparse long-term full backups alongside short-term daily recovery points. A typical policy: 14 daily recovery points, 4 weekly, 12 monthly and a few yearly full backups. Two points determine the capacity calculation:
- GFS points are full backups; they aren’t stored as incrementals. Without block cloning, each one takes up the size of a full backup.
- Keeping long-term points in the primary repository isn’t mandatory. Monthly and yearly points are usually copied to a secondary repository, object storage or tape. The choice of media for this tier is covered in the LTO tape vs. object storage comparison article.
The systems team doesn’t choose retention periods on its own; the corporate retention policy and regulations set them. The “let’s keep it seven years, just in case” approach is also a mistake: keeping data longer than necessary can cause problems both in disk cost and in personal data retention obligations.
Worked example table
The worked example below is based on assumptions: 10 TB of used source data, 2:1 data reduction, a 3% daily change rate, 14 recovery points, and GFS of 4 weekly + 12 monthly + 3 yearly. In the block cloning scenario, each GFS full backup is assumed to take up only the unique blocks changed since the previous retained full (weekly 0.5 TB, monthly 1.2 TB, yearly 2.5 TB; reduced size).
| Item | Calculation | No block cloning | With block cloning |
|---|---|---|---|
| Full backup | 10 TB ÷ 2 | 5 TB | 5 TB |
| Incremental (daily) | 10 TB × 3% ÷ 2 | 0.15 TB | 0.15 TB |
| 14 recovery points | 5 + 13 × 0,15 | 6.95 TB | 6.95 TB |
| 4 weekly fulls | 4 × 5 TB; cloned: 4 × 0.5 TB | 20 TB | 2 TB |
| 12 monthly fulls | 12 × 5 TB; cloned: 12 × 1.2 TB | 60 TB | 14.4 TB |
| 3 yearly fulls | 3 × 5 TB; cloned: 3 × 2.5 TB | 15 TB | 7.5 TB |
| Subtotal | 101.95 TB | 30.85 TB | |
| 3-year growth (15% per year) | × 1,52 | ~155 TB | ~47 TB |
In Veeam, the same full backup can carry more than one GFS flag (for example, the last weekly full of the month is also the monthly point); because of this overlap, real consumption may come out slightly lower than the table value. The table is an upper estimate that stays on the safe side. Working space for merges and temporary operations is added to the result. Define an internal upper threshold for repository usage (for example 80% full) and set the gross capacity to buy according to this threshold. RAID or erasure coding protection overhead must also be added to gross capacity: in a 12-disk RAID 6, the capacity of two disks is used for parity.
The table shows two things. In policies that include GFS, block cloning reduces capacity many times over. In an environment without cloning, the most effective saving is moving long-term points out of the primary repository. In the example, the difference is between about 102 TB and 31 TB.
Block cloning and deduplication
ReFS on Windows Server and reflink-enabled XFS on Linux can reference the same data in a new file’s blocks without physically copying it. Veeam uses this feature under the name “Fast Clone”: creating synthetic full backups and merging incrementals becomes a metadata operation instead of moving data. Merges that took hours drop to minutes, and space is saved too.
A common mistake here is formatting the repository volume with the default 4 KB cluster size and noticing later; the cluster size can’t be changed afterwards, and the volume has to be reformatted. Examples of formatting with the 64 KB cluster size Veeam recommends for ReFS and with reflink for XFS:
# Windows Server – ReFS, 64 KB cluster
Format-Volume -DriveLetter E -FileSystem ReFS -AllocationUnitSize 65536 -NewFileSystemLabel "VeeamRepo"
fsutil fsinfo refsinfo E:
# Linux – XFS with reflink enabled
mkfs.xfs -b size=4096 -m reflink=1,crc=1 /dev/sdb
xfs_info /mnt/veeamrepo | grep reflink
Limits of block cloning:
- The saving applies only within the same volume. When backups are copied or moved to another repository, the blocks are physically written again.
- Active full backups aren’t cloned; each active full physically takes up full size.
- The sum of file sizes doesn’t show real usage. On Linux,
ducounts shared blocks separately in each file; for the correct value, look at the volume’s free space withdf. On Windows, too, the volume’s free space should be used rather than file properties.
Hardware deduplication appliances reduce repeated blocks in backup files globally. They can provide significant savings in GFS policies with long retention; but the cost of rebuilding the data (rehydration) during restore lengthens the RTO. Writing is fast, reading is slow. Keeping the latest recovery points that need fast restores on a classic repository and long-term points on a deduplication appliance is a common tiering method. Before adding operating system-level deduplication on top of backup files, check the backup software vendor’s support position.
Growth headroom and monitoring
A capacity calculation is correct on the day of purchase, not six months later. Compare the plan with actual usage regularly:
- Record repository fill level, full and incremental backup sizes and the actual data reduction ratio every month.
- Update the calculation when a new system is commissioned, especially when a large database or file server is added.
- Define a monitoring rule that raises an alert before the fill threshold is reached. A full repository means no backups from that night on.
- If multiple repositories are used, monitor the placement policy and extent capacities together; scalable repository design is described in the Veeam proxy and repository architecture article.
Calculation checklist
- Was the source size taken from used blocks?
- Was the change rate measured by system type, including peak days?
- Does the data reduction ratio come from real job history?
- Has it been decided which repository will hold the GFS points?
- Does the file system support block cloning, and was it formatted correctly?
- Were the second and third copies (offsite, immutable copy) calculated separately?
- Were RAID/erasure coding overhead, working space and the fill threshold added to gross capacity?
For copy count and location requirements, see the 3-2-1-1-0 backup rule article, and for a free start in small environments, the Veeam Community Edition article. You can see how we work when choosing and sizing repository hardware on our hardware consulting and backup and disaster recovery pages. How ReFS block cloning works is described in Microsoft’s ReFS block cloning document.
Frequently asked questions
How many times the source data does a backup repository need?
There is no fixed multiplier. Depending on the change rate, data reduction ratio, GFS retention policy and block cloning support, the capacity needed for the same source data can differ several times over. Do a worked calculation with these variables.
How is the daily change rate measured?
The most reliable method is to look at the incremental backup sizes of existing backup jobs over the last few weeks. If there are no backups, run a short trial job. Disk write counters overstate the change rate because they also count the same block being written again.
How much space does ReFS or XFS block cloning save?
Because synthetic full backups and merge operations create block references instead of copying data, GFS full backups take up only as much space as the changed unique blocks. The saving depends on the retention policy and applies only within the same volume.
Should GFS backups be kept in the primary repository?
It isn’t mandatory. Monthly and yearly points are usually copied to a secondary repository, object storage or tape. The fast primary repository is reserved for short-term restores, and capacity cost drops.