tv42
commented
3 years ago ARC is still patented, isn't it? W-TinyLFU is probably the worthy pick.
https://arxiv.org/abs/1512.00727
https://9vx.org/post/on-window-tinylfu/
https://github.com/ben-manes/caffeine/wiki/Efficiency
the initial implementation should borrow heavily from the ARC, using ghost lists on both an MRU & MFU cache to determine how perf would be if they were the only cache, and use collected statistics on these to decide how much actual data each should be caching.