Re: Performance tweaks for large caches

On Thu, 13 Jun 1996, Tai Jin wrote:

> I would expect that an LFU algorithm is better for objects which I
> expect to stick around in the cache for a while. Perhaps you can
> implement a hybrid -- LRU for short-lived objects and LFU for
> long-lived objects. Increased frequency of access might bump an
> object from LRU to LFU and vice-versa for decreased frequency.

You are probably right, and it should not be that hard to
implement, but I am not sure that we gain that much in cache
ratio, compared with what you loose in CPU usage and
complexity (== needs more time to implement).

---
Henrik Nordstrom