Projects

Backup cost engineering: from ~74 to ~20 CAD a month

Designing off-site backup lifecycle for terabytes of family data under a fixed budget, without giving up restorability.

~74 → ~20CAD / month at 2.4 TB
50 CADhard monthly budget, with alerts
0age-based deletes on deduplicated data
  • Cloud cost
  • Backups
  • Azure
  • Lifecycle management
  • FinOps

The problem

Photo and file libraries grow into terabytes. Kept in the default storage tier, the projected off-site backup bill was about 74 CAD a month, well over a 50 CAD budget. The constraint was clear: cut the cost, but never lose data.

What drives the cost

  • Mostly media. Nearly all of the volume is photos and files that rarely change after upload.
  • Rebuildable data was costing the most. Metrics and logs were two thirds of the backup data, rewritten every day, and they can be rebuilt.
  • The repository layout limits the options. Deduplicated backup repositories keep blocks written by the first backup for as long as the data is unchanged. Deleting blobs by age would silently corrupt every backup.

The design

  1. Stop backing up what can be rebuilt. Metrics, logs and scratch volumes are excluded. Databases are excluded at file level, because they’re covered by point-in-time archiving and logical dumps.
  2. Tier, don’t delete. Data blocks of the big libraries move to the Cold tier after 30 days. Cold is about 4.4× cheaper than Hot and still online, so backups and restores work unchanged. Indexes and metadata stay Hot.
  3. Let the tools delete their own data. Retention lives in the backup tools: 7 daily, 5 weekly and 6 monthly copies, and 14 days of database history. Storage rules only expire simple dated files.
  4. Keep the undo button. Blob versioning and soft delete stay on. They cost little and undo mistakes made by people, tools or rules.
  5. Watch it.
    • A monthly budget with e-mails at 15, 30 and 45 CAD, and a forecast warning before 50.
    • A capacity alert.
    • A one-command usage report.
    • A five-minute monthly review.

The numbers

Scenario (2.4 TB of media) Monthly cost
Everything in the default (Hot) tier ~74 CAD
Rebuildable data excluded, media in Cold after 30 days ~20 CAD
Headroom inside a 50 CAD budget ~6 TB

The trade-off we accepted: a full disaster restore would pay a one-off retrieval and egress cost of about 360 CAD. Partial restores, the common case, cost cents.

Results

  • Under budget with room to grow, and no data protection given up.
  • Cost is now an alerted signal with a written response plan, not a surprise on the invoice.