Juiceman wrote:
>>> the 50:50
>>> split seems pretty arbitrary; it should be configurable.
>> Why, so each user can choose a different arbitrary value instead of
>> everyone using the same arbitrary value?
>>
> 
> Well, I for one, would like to be a "deep cache" for long term
> storage, others like Bob would like to cache as much as possible
> immediately.  I think Freenet could benefit from this.

Would it be possible to dynamically balance the amount of space 
allocated to the cache and store? Here's what I have in mind: when the 
cache is below its quota and a key is cached, remove the least recently 
used key from the store instead of the cache. And vice versa: when the 
store is below its quota and a key is stored, remove the least recently 
used key from the cache. That way we don't waste any space while the 
store is filling up (the cache uses the free space), and if we later 
change the quotas it will gradually move to the new quotas instead of 
having to resize the whole thing at once.

On a related note, is there any evidence that LRU performs better than, 
say, random replacement for Freenet's purposes? IIRC there's some 
evidence that LRU in a distributed cache produces too many copies of 
popular data:

http://www.cs.princeton.edu/~qlv/download/searchp2p_ics02.pdf
http://gnunet.org/papers/p2pmulti.pdf

Cheers,
Michael

Reply via email to