The multi-generational LRU

The mult-generational LRU seems very close to the scheme that I developed (along with David Greenman) back in the early 1990s. My scheme was not very scalable, but allowed running X windows in about 4MB on FreeBSD (actually usefully at 8MB.) The scalability was good enough into the 100's of MB range, but beyond that size, it would need some algorithmic help. It sometimes made a substantial difference, sometimes not much at all. I forget what we called it, but it had both an active/inactive along with accumulating usage counts on a page basis. The pages would cycle though the queues and increment (or decrement) usage counts as appropriate. Pages could become fairly sticky or be made eligible for quick reuse. I truly do not know if FreeBSD still uses the scheme.