That could work efficiently in a sampled caching policy. For example, Hyperbolic [1] maintains the frequency and insertion time on each entry and evicts the ent

That could work efficiently in a sampled caching policy. For example, Hyperbolic [1] maintains the frequency and insertion time on each entry and evicts the entry in the sample with the least utility value of frequency / (currentTime - insertionTime). For yours, you could easily add it to the SampledPolicy in this simulator [2] and run it against a few traces to see how it compares. [1] https://www.usenix.org/system/files/conference/atc17/atc17-blankstein.pdf https://www.usenix.org/system/files/conference/atc17/atc17-b... [2] https://github.com/ben-manes/caffeine/wiki/Simulator https://github.com/ben-manes/caffeine/wiki/Simulator