Amey-Thakur Media storage tiering
Media accumulates and is read on a sharply decaying curve: most files are accessed heavily for days and almost never afterwards. Paying hot storage prices for cold data forever is the default and is usually the largest avoidable cost in a media product.
Method
- Measure the access curve before designing tiers. How access decays with age determines the transition points, and guessing produces either retrieval fees or no savings.
- Transition by age and access, automatically. Lifecycle rules that move objects without application involvement, since manual tiering does not happen.
- Understand the retrieval cost and latency of each tier. Archive tiers are dramatically cheaper to store and expensive and slow to read, which is fine for backups and wrong for anything a user might open.
- Keep derived assets hot even when originals are cold. Thumbnails and streaming renditions are what get served, so they stay accessible while the source archives (see thumbnail-generation).
- Check minimum duration charges. Cold tiers bill a minimum retention, so moving short-lived objects there costs more than leaving them.
- Delete rather than archive where policy allows. The cheapest storage is none, and retention policy should be set deliberately (see right-to-erasure).
- Monitor cost per tier and per feature. Aggregate storage cost hides which feature is responsible for the growth.
Boundaries
Tiering optimises cost against access latency; it cannot make cold data fast. Retrieval charges can exceed the storage savings if the access pattern was misjudged. Legal retention requirements may dictate storage class and duration regardless of cost (see data-classification).