Layer Reuse Ratio Across an Image Set

1. Background Knowledge

In container image registries, images are stored as a stack of layers, each identified by a unique content-addressable digest (a hash). When multiple images share the same layer, the registry stores that layer only once on disk, but the logical size of each image still includes it. This is the foundation of layer reuse: the physical storage footprint is the union of all distinct layers, while the logical footprint is the sum across all images.

The naive total is the sum of the sizes of every layer in every image, counting shared layers multiple times. The deduplicated total is the sum of sizes of each unique digest, counted exactly once. The reuse ratio quantifies the storage savings from sharing: it is the fraction of the naive total that is eliminated by deduplication. Mathematically, if $N$ is the naive total and $D$ is the deduplicated total, the ratio is $1 - \frac{D}{N}$, with a special case of $0.0$ when $N = 0$ to avoid division by zero.

This metric is useful in CI/CD pipelines and registry management to understand how much disk space is saved by multi-stage builds, base image sharing, and other layer-reuse strategies.

2. Algorithm Approach

This is a set-based deduplication problem. The core idea is:

1. Compute the naive total by summing all (digest, size) pairs across all images.
2. Compute the deduplicated total by collecting all unique digests and summing their sizes once each.
3. Apply the ratio formula with the zero-division guard.

The key data structure is a dictionary (or set) keyed by digest, which naturally handles deduplication in $O(1)$ average lookup time.

3. Step-by-Step Strategy