Scenario · Last verified 2026-08-25
Backup and archive
Retention, rare restores, and the request mix of a backup bucket.
| Metric | Value | Unit | Source type | Derivation | Link |
|---|---|---|---|---|---|
| Retained unique bytes | 1500 | GB | Author assumption assumption | 1 TB protected source plus thirty days of three-percent daily increment, overlapping blocks counted once. Snapshot chains can exceed this. | — |
| Restore and offsite pull | 50 | GB | Author assumption assumption | One fifty-gigabyte restore test a month. Production DR drills that copy a full TB are a different scenario. | — |
| Backup PUTs | 120000 | request | Author assumption assumption | Incremental object PUTs and multipart parts for daily backups, on the order of thousands per day. | — |
| Verify and restore GETs | 8000 | request | Author assumption assumption | Inventory and restore-test reads, not a public download pattern. | — |
Adjust the assumptions
| Vendor | Monthly total at the hub usage |
|---|---|
| Cloudflare | $22.35 |
| Azure | $27.51 |
| Google Cloud | $29.90 |
| AWS | $34.98 |
Breakeven: cross-region-read-vs-multi-region-replica
Assumptions and limits
Backups are a storage-and-PUT problem until the day you restore. Fifty gigabytes of restore traffic stays inside common internet grants; a full terabyte restore does not, and that is when R2's zero egress or a same-region restore path matters. Glacier and Archive classes need duration and retrieve-fee assumptions that this snapshot does not carry, so the ranking here is Standard versus R2, not cold storage. Cross-region copy of every daily snapshot doubles capacity without improving RTO unless you actually test the copy. Turn free-tier off if another workload already consumed the grant. Capacity meters: https://aws.amazon.com/s3/pricing/ https://developers.cloudflare.com/r2/pricing/