Estimate ECR storage (GB-month) from images and retention

Registry storage cost is usually driven by how many images you keep and how large they are. You can get a good estimate with a simple model (average size × images retained), then refine it using push frequency and retention windows.

Step 1: define repo classes (do not use one average)

CI artifacts: many tags/day, short retention (should be cheap if retention works).
Releases: fewer tags, longer retention (rollback needs).
Base images: few images but large and shared across many services.

Fast model (good for budgeting)

Stored GB-month ~= avg image size (GB) × retained images (count)
Retained images per repo ~= tags retained × (variants per tag)
If you store multi-arch variants separately, variants per tag can be 2 (amd64 + arm64) or more.

Better model (push frequency + retention days)

This model is useful for repos where CI pushes constantly.

Pushes/day = N
Retention days = D
Retained images ~= N × D (per repo or repo class)
Stored GB ~= retained images × avg image size

Worked example

Avg image size: 0.8 GB
CI pushes: 40 images/day
Retention: 14 days
Retained images ~= 40 × 14 = 560 images
Stored GB ~= 560 × 0.8 ~= 448 GB-month (order-of-magnitude)

Common multipliers (what makes storage explode)

No retention: “keep forever” means storage grows without bound.
Multi-arch: storing multiple variants can multiply retained GB.
Large base images: shared images can dominate storage even if app images are small.
Environment duplication: separate repos/tags for prod/staging/dev with independent retention.

What to measure first (so your estimate converges)

Average image size by repo class (CI vs release vs base images).
Push frequency per repo (images/day) for the top 10 repos by activity.
Retention rules (days or tags) and whether untagged images are deleted.
Whether multi-arch variants are retained equally (amd64 + arm64 doubles storage if both are kept).

Turn storage into cost

Use AWS ECR Cost Calculator with your stored GB-month estimate and your effective $/GB-month.

Validation checklist

Pick 3 representative repos and measure average image size and retained tags.
Validate retention policies actually delete old tags (and untagged images).
Validate multi-arch behavior (are variants stored separately and retained equally?).
After changes, compare predicted GB-month to registry-reported storage.