Open bdupras opened 8 years ago
Oh - and by the way, this is work in progress. This version of the CuckooFilter doesn't yet function. When I get it functional (soon), I'll probably break it out from the guava fork into its own repo.
@bdupras , my past experience is 4 cells per bucket ensures pretty good load factor --- which is 95%+ with 500 retries on insert; Using 8 cells per bucket will get even higher load factor (e.g., close to 99%) with even less retires, but at the cost of doubling the false positive rate. I would say 4 or 8 is a good value to use in practice
The current methods for determining bucket size (number of entries per bucket) and load factor are based on prose in section 5.1 of Cuckoo Filter: Practically Better Than Bloom.
It would be better to have true calculations instead of the current conditional logic with hard-coded values.
Heya @apc999 - do you have any recommendations here?
// cc @beala