LRU cache

Least Recently Used caching is one of the oldest techniques in computing. When a new item needs to be stored but the cache is at capacity, the item that was accessed furthest back in time is removed. The result is that frequently accessed items stay in the cache, and cold items leave on their own.

Under the hood, an LRU cache tracks the order in which items were last touched. Every read or write moves that item to the front of a recency list; when space is needed, the item at the back, the one untouched longest, is dropped. A well-built implementation does this in constant time with a hash map plus a linked list, so tracking recency adds no real overhead. LRU is one policy among several: LFU evicts the least frequently used, FIFO evicts the oldest written regardless of use, and each suits a different access pattern. LRU wins when recent use predicts near-future use, which is exactly how an editor returns to the same clips across a session.

For video editors, LRU means your cache never manually fills up or stops working. You set a size budget (for example 50 GB), and the system handles the cleanup. Pinned items are excluded from eviction, so critical project files can be protected manually.

What makes LRU ideal here is that eviction is cheap and fully reversible. The proxies on your SSD are disposable copies; the authoritative originals live in cloud storage. So when the cache evicts a proxy you have not touched in weeks, nothing is lost, the moment you reopen that project the proxy is regenerated or the original streamed on demand. That safety net is why an aggressive, automatic cache works for editing: you get the speed of local media without ever babysitting disk space or fearing data loss.