Simple lru cache?

TeaAndBikkie teaandbikkie at aol.com
Sat Oct 5 17:06:52 CEST 2002


>From: Ype Kingma ykingma at accessforall.nl 

>I'd expect in the order of 10000 items cached of which:
>20% would be used around 4 times,
>20% around 3 times,
>20% around 2 times,
>40% used once only.
>

Perhaps you could improve the performance of the LRU cache by adding a small
complexity to preferentially keep those items most reused, in combination with
recent items.

Say keep a count of how many hits for each item, then when removing items from
the cache, keep an older more-used item in preference to a newer one-hit-wonder
item (unless the older one is older than an agreed expiry time, to prevent a
senile cache from piling up). This might prevent the cache getting filled with
recent stuff that is not needed again...

I suspect that generally this has some disadvantages, but in your case it might
work well.

-- Misha




More information about the Python-list mailing list