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.
A deduplication ratio such as 10:1 divides the amount of data the backups represent by the amount physically stored. It says how much repetition the store collapsed into single copies. It does not say what the storage costs, because the top number grows with every restore point you keep while the bottom number barely moves. To budget, convert any ratio into GB stored per GB of source data.
What are the two numbers behind a ratio?
A ratio has a top and a bottom, and reports are often vague about both.
- Logical (protected) size: how big the data would be if every restore point were stored as a full, separate copy. Thirty daily restore points of an 800 GB server is 30 × 800 GB = 24,000 GB logical.
- Physical (stored) size: what the store actually holds after it keeps each unique piece once, and usually after compression.
Ratio = logical ÷ physical. A report of 10:1 means ten units of protected data per one unit stored.
The same arithmetic converts between a ratio and a "savings" percentage. Savings = 1 − (1 ÷ ratio).
| Ratio | Space saved | Stored per 100 GB protected |
|---|---|---|
| 2:1 | 50% | 50 GB |
| 5:1 | 80% | 20 GB |
| 10:1 | 90% | 10 GB |
| 20:1 | 95% | 5 GB |
| 50:1 | 98% | 2 GB |
Microsoft's Data Deduplication overview uses the same equivalence: "optimization rates of up to 95%, or a 20x reduction in storage utilization". Note what moves between rows. Going from 10:1 to 20:1 halves the stored data, but going from 20:1 to 50:1 saves only another 3 points. Higher ratios give diminishing savings, so a 50:1 headline may be only 3 percentage points better than 20:1.
Why does the ratio climb the longer you keep backups?
Because each extra restore point adds to the top number by the full size of the source, and to the bottom number only by what changed.
Take a server of 800 GB. After the first full copy, assume each day adds 13 GB of new stored data (the arithmetic behind this figure is set out in the related piece on why a backup store is bigger than its data). The first copy is stored as 650 GB and is the oldest restore point, so N restore points hold N − 1 days of change.
| Restore points kept | Logical (GB) | Stored (GB) | Ratio | Stored per GB protected |
|---|---|---|---|---|
| 7 | 7 × 800 = 5,600 | 650 + 6 × 13 = 728 | 5,600 ÷ 728 = 7.7:1 | 728 ÷ 800 = 0.91 |
| 30 | 30 × 800 = 24,000 | 650 + 29 × 13 = 1,027 | 24,000 ÷ 1,027 = 23.4:1 | 1,027 ÷ 800 = 1.28 |
| 60 | 60 × 800 = 48,000 | 650 + 59 × 13 = 1,417 | 48,000 ÷ 1,417 = 33.9:1 | 1,417 ÷ 800 = 1.77 |
The ratio more than quadrupled, from 7.7:1 to 33.9:1, while the store grew from 728 GB to 1,417 GB. The ratio looks better each time you keep more, yet the bill rises each time. A provider that lets you extend retention will also show you a ratio that improves.
Two backup setups can show the same ratio and cost very different amounts. A ratio can never be compared across two sites unless the retention and the source size are the same.
Do restic and Borg print a deduplication ratio?
Neither prints a ratio. They print sizes, and you divide.
restic reports sizes through restic stats, whose --mode flag selects what is counted. The manual describes the modes as:
restore-size(the default): "Counts the size of the restored files."files-by-contents: "Counts total size of unique files, where a file is considered unique if it has unique contents."raw-data: "Counts the size of blobs in the repository, regardless of how many files reference them."blobs-per-file: "A combination of files-by-contents and raw-data."
A usable restic ratio is therefore restore-size for the snapshots you care about against raw-data for the repository. The manual page gives one-line definitions only. It does not state whether raw-data is counted before or after compression.
Borg prints three columns. In borg info, the sizes are labelled Original size, Compressed size and Deduplicated size, in two rows: "This archive" and "All archives". The documentation warns that "the deduplicated sizes of the individual archives do not add up to the deduplicated size of the repository ('all archives'), because the two mean different things". "This archive" deduplicated size is "amount of data stored ONLY for this archive". "All archives" deduplicated size is "amount of data stored in the repository". The page does not define Original size or Compressed size beyond that.
Borg's layout is useful because compression and deduplication are shown as separate steps. A report that shows one combined ratio does not tell you which step produced it, so ask whether the figure includes compression.
What does Microsoft's "SavingsRate" mean?
Read its definition before comparing it with a ratio. Microsoft's page "Running Data Deduplication" says Get-DedupStatus has two fields relevant to the optimisation rate. It defines OptimizedFilesSavingsRate as "space used by optimized files after optimization / logical size of optimized files", and SavingsRate as "space used by optimized files after optimization / total logical size of the optimization".
Written as a fraction of used over logical, that formula describes the share still stored, not the share saved. The overview page calls 95% "a 20x reduction", which treats the figure as a saving. The two pages are not aligned on direction, so do not assume which way a given tool reports. Check against sizes you can see: if the stored size is a small part of the logical size and the field shows a small number, it is reporting the fraction stored. If it shows a large number, it is reporting savings.
Windows Data Deduplication also works on one volume at a time, so its savings describe the volume, not a multi-month retention history in a remote store.
What figure should a budget use instead?
Use stored GB per GB of source data, at a stated retention. It is a plain multiplier: 1.28 means every 1 GB of source data costs 1.28 GB of stored space at that retention. It rises when retention is lengthened, when data churns more, or when more of the data will not compress. It can be checked against the storage's own usage figure each month. A ratio cannot be checked that way, because the top number is never physically stored anywhere.
Which quote is cheaper?
Example (illustrative): a finance reader receives two quotes for the same 800 GB server from two providers using the same style of store, quote A and quote B. Quote A keeps 7 restore points and reports a ratio of 7.7:1. Quote B keeps 60 and reports 33.9:1. A reader who picks the larger ratio has picked the larger store.
Convert both to GB stored per GB protected: A is 728 ÷ 800 = 0.91. B is 1,417 ÷ 800 = 1.77. At a rate of R per GB-month, A is 728 × R and B is 1,417 × R. B costs 1,417 ÷ 728 = 1.95 times as much, and it also keeps more than eight times as many restore points. The 7.7:1 against 33.9:1 headline hides both facts. The right comparison is cost per restore point kept, or stored GB for the retention you actually need.
How do I check this on my own system or with a provider?
- Ask for the two raw numbers: logical size and stored size, in GB with the unit stated (GB or GiB), and the date they were measured.
- Ask what is in the top number: source data once, or every restore point counted in full.
- Ask whether compression is included in the ratio, or counted separately.
- Ask for the retention behind the figure. A ratio without it cannot be compared.
- Do the division yourself: stored ÷ source gives stored per GB protected.
On a restic repository, these are the two read-only commands from the manual:
restic stats --mode raw-data
restic stats --mode restore-size latest
On Borg:
borg info
Take the "All archives" deduplicated figure as the physical size. Compare the physical size with the storage's own usage figure for the folder or bucket, because reported and billed sizes can differ.
Read next
Related articles you may want next.
- Backup & RecoveryBackup storage is bigger than the data it protects: where the extra space goesWhy backup storage can exceed the server it protects: retention, changed data, compression, unreclaimed space and chunk rounding, with arithmetic.Read article
- Backup & RecoverySize Backup Retention Using Measured Storage GrowthBuild a capacity forecast from stored growth, copy formats and retention dependencies, with explicit headroom, recovery needs and uncertainty.Read article
- Backup & RecoveryFixed or variable block deduplication: why it changes how much backup storage you useHow fixed-size and variable-size block deduplication differ, what an insert does to each, how chunk size drives index size, and what to ask a provider.Read article
Risks and limits
- These commands have not been run or lab-tested. They are taken from restic, Borg and Microsoft documentation.
- The worked figures are made up and use round assumptions: unique daily changes of 13 GB stored, a 650 GB first copy and no clean-up overhead.
- The restic manual does not say whether
raw-datais counted before or after compression. Compare it with the real storage usage rather than assume. - Microsoft's two pages disagree in wording on the direction of
SavingsRate, as described above. Both definitions are quoted word for word, and neither is treated as the correct one. - A ratio from one product or period cannot be compared with another. Borg 2.x and other restic versions may print different fields.
- Whether a provider counts shared data across customers, or bills on protected size rather than stored size, is a contract matter that no ratio shows.
Sources
- Data Deduplication overview: Microsoft Learn, page date 2025-06-20, accessed 17 November 2026
- Understanding Data Deduplication: Microsoft Learn, page date 2022-02-18, updated 2025-09-22, accessed 17 November 2026
- Running Data Deduplication: Microsoft Learn, page date 2022-02-18, updated 2025-09-15, accessed 17 November 2026
- restic-stats(1), v0.19.1: restic project, accessed 17 November 2026
- borg info: Borg 1.4.5 documentation, no page date, accessed 17 November 2026
- borg create: Borg 1.4.5 documentation, no page date, accessed 17 November 2026
This article was originally published on .