I call it a probabilistic cache because it relies on probabilistic features of its underlying Cuckoo Filter. There is a high chance that most recently/frequently used items will remain in the cache and least recently/frequently used items are either not cached at all or removed lazily to give space to recently/frequently accessed items.
CuckooCache uses a cuckoo filter to keep track of accessed/cached items, and uses an ets table to store cached items.
Every time an item is accessed, its key is inserted in the cuckoo filter. An item is cached in the ets table only if it already exists in the cuckoo filter. In other words items that are accessed once and never used again in a sliding window of requests (one-hit-wonders), will not be cached.
When an item is added to the cuckoo filter, and filter capacity is full, another random item gets removed from the filter, and removed item is also removed from the ets cache if it exists in the cache and no longer exists in the filter.
One interesting feature of a cuckoo filter is that it can contain duplicate elements (up to bucket_size * 2 times). This feature gives frequently accessed elements more chance to stay in the filter as long as they are accessed frequently enough.
Feed backs are welcome!