Parallelism executes multiple transfers concurrently. Docker Engine downloads image layers at the same time—three by default—so splitting large data across layers can improve pull performance. section 17.4
Apply parallel layer transfers when optimizing large Docker images:
- Prefer multiple layers for large transfer-heavy images, while recognizing that explicitly listing every file is more brittle. section 17.4
ML dataset image
A one-layer target copies days 70–79 with a wildcard; a multi-layer target copies each data file separately. Pulling the multi-layer image can be substantially faster on a new machine.
