Keeping every website backup in S3 Standard is simple. Moving older copies to a Glacier class with a lifecycle rule is cheaper on paper. This article works through one example with a small model so you can see where the saving comes from, which inputs matter and what the figures leave out. For the general rules on RPO, RTO, retention and restore testing, read the backup strategy guide. If you are wondering whether a second provider belongs in the picture, see when adding Cloudflare R2 reduces backup costs.
The model behind the numbers
The AWS S3 backup calculator and the examples below use the same steady-state model:
- Each backup is a full, independent set of G GB, created every F days.
- Each set spends S days in S3 Standard, then A days in the chosen Glacier class. A is at least 90.
- Approximate monthly storage cost = (G / F) × (S × Standard rate + A × archive rate).
This is a smooth daily occupancy estimate, not an AWS invoice. Backup size is fixed, with no incrementals, deduplication or growth. Those simplifications keep the comparison fair, but they also mean the output is a planning figure rather than a bill. S and A are the periods you configure, and the lifecycle template the calculator produces expires objects at S + A for every frequency. Frequency changes only how many sets exist at once, through G / F; it does not change how long each set lives.
Three classes, three retrieval behaviours
AWS describes its classes in the storage class overview. For backups the practical differences are:
- S3 Standard. Immediate access and the highest storage rate of the three.
- S3 Glacier Instant Retrieval. Objects stay immediately accessible, with a lower storage rate and a retrieval charge per GB read.
- S3 Glacier Flexible Retrieval. Objects must be restored before you can read them. How long that takes depends on the retrieval tier you request and on available capacity. The restored copy is a temporary object billed at Standard storage rates in addition to the archived one.
“Glacier is slow” is therefore not a useful summary. Instant Retrieval reads like Standard; Flexible Retrieval adds a restore step. In both cases a full site recovery still needs time to transfer the data and validate it. S3 Glacier Deep Archive also exists, but the calculator does not model it and this article does not use it.
Rates used in the example
| Class | Rate (USD per GB-month) |
|---|---|
| S3 Standard, first 50 TB | 0.025 |
| S3 Glacier Instant Retrieval | 0.005 |
| S3 Glacier Flexible Retrieval | 0.0045 |
These come from the public AWS price list for ap-southeast-1 (Singapore), effective 2026-10-01 and reviewed on 2026-11-06. The calculator reads a checked-in snapshot of that file, and a build refresh can update the snapshot. Check the rates and date displayed in the tool before you rely on any figure here.
A worked example
Take the calculator defaults: G = 100 GB, F = 1 day, S = 1 day in Standard, A = 90 days in Glacier Instant Retrieval. Each object lives 91 days in total, so at steady state about 91 sets exist at once.
Standard only. Keep every copy in Standard for the full 91 days:
100 GB × 91 days × $0.025 = $227.50 per month.
Lifecycle to Glacier Instant Retrieval.
- Standard: 100 GB × 1 day × $0.025 = $2.50
- Archive: 100 GB × 90 days × $0.005 = $45.00
- Total: $47.50 per month
The storage difference is $180.00 per month. Treat that as an illustrative, storage-only figure for a fixed 100 GB daily backup in one region, before the exclusions listed below. It is not a savings guarantee, and it includes no restore activity. Flexible Retrieval lowers the archive line a little further, at the cost of the restore step.
Reading the calculator inputs correctly
Two inputs are easy to misread.
“Days in Glacier” is archive residence, not total age. The archive input counts days after the transition. With the defaults, total object age is 1 + 90 = 91 days. If you enter 90 and expect objects to disappear on day 90, your policy and your estimate will disagree.
Expiration is total object age. An S3 lifecycle expiration value counts from object creation, not from the transition. With the defaults, S + A = 91 days, and that is the expiration age to set. The estimate prices the S and A you configure, nothing more.
Frequency changes occupancy, not object age. The lifecycle template expires objects at S + A whatever F you choose. Take F = 30, S = 1 and A = 90 with 100 GB in Glacier Instant Retrieval: the template sets the transition at day 1 and expiration at day 91, each set spends 90 days in the archive class, and the estimate is (100 / 30) × (1 × $0.025 + 90 × $0.005), about $1.5833 per month. A set is created every 30 days, so about three sets exist at once instead of 91. The 30-day interval changes G / F only; S and A stay what you entered. S3 rounds age-based actions to the next midnight UTC and runs them asynchronously, so expiration at day 91 is the age in the rule, not an exact deletion moment.
Two further points about the tool:
- There is no Standard-only option in the class dropdown, and the archive input has a 90-day floor. Do not try to emulate Standard-only by setting Standard days to the full retention. Compute the baseline by hand instead. With T = S + A as the expiration age, the Standard-only baseline is (G / F) × T × Standard rate, and the calculator’s lifecycle estimate is (G / F) × (S × Standard rate + A × archive rate). Both use the same smooth daily occupancy as the calculator, so they are planning figures, not invoices. In the daily example T = 91, which is why the baseline above is 100 GB × 91 days × $0.025. At F = 30 the same baseline is (100 / 30) × 91 × $0.025, about $7.5833, against the $1.5833 lifecycle estimate.
- The rotation comparison changes F while keeping the selected class. It compares backup frequencies, not Standard-only against lifecycle. A longer F lowers cost and also removes restore points, so it changes recovery coverage, not just price.
The 90-day minimum and the common trap
S3 pricing states that both Glacier Instant Retrieval and Glacier Flexible Retrieval carry a 90-day minimum storage duration, counted after the transition. Deleting, overwriting or transitioning an object earlier incurs a pro-rated charge for the remaining days. Instant Retrieval also has a 128 KB minimum billable object size, and Flexible Retrieval adds a small per-object metadata overhead.
The trap looks like this: 30 days in Standard, then a rule expiring objects at 90 days total. Objects only spend 60 days in the archive class, so the remaining 30 days are billed as an early deletion. The calculator’s archive input sidesteps this particular trap by treating A as archive residence with a 90-day floor, which is why the default produces a 91-day total rather than 90. That floor covers the minimum duration only; it does not review the rest of your lifecycle rule for you.
Lifecycle details that change the outcome
The lifecycle transition considerations page notes a few points that matter for backup buckets:
- Standard to Glacier transitions are supported, but each object transition is a request with its own charge. A backup made of thousands of small files costs more to transition than one archive of the same size.
- Current lifecycle rules do not transition objects smaller than 128 KB by default. Some pre-existing configurations behave differently, so review existing rules rather than assume.
- Lifecycle expiry deletes objects by age. If your backup tool writes incremental chains or keeps a catalog, expiring a parent object can break every dependent backup. The model here assumes full, independent sets precisely to avoid that.
What the estimate leaves out
The calculator prices storage only. Not included: PUT, GET and lifecycle transition requests; retrieval charges; data transfer out of AWS; metadata and minimum-size overhead; early deletion charges; taxes and support plans. For a backup that is rarely restored, storage dominates. For a backup you restore weekly to a staging server, retrieval and transfer can rival the storage line, and that is the scenario the next article examines.
Next step
- Pick G from your current backup size, F from your recovery point objective and S from how far back routine restores usually reach.
- Set A to the archive residence you want, 90 days or more. Set expiration at S + A, which is the age the calculator’s lifecycle template sets for every frequency.
- Run the inputs through the AWS S3 backup calculator, confirm the displayed rates and date, and compute the Standard-only baseline by hand at the same S + A age, so both policies are priced on the same residence.
- Test a restore from the archive class independently, so the retrieval behaviour is known before you need it.
If you want a second pair of eyes on the inputs for a specific site, the web monitoring service page describes what we look at. Otherwise, continue to when adding Cloudflare R2 reduces backup costs.