Similar to a Bloom filter, a Cuckoo filter provides a space-efficient data structure designed to answer approximate set-membership queries (e.g. "is item x contained in this set?") Unlike standard Bloom filters, however, Cuckoo filters support deletion. Likewise, Cuckoo filters are more optimal than Bloom variants which support deletion, such as counting Bloom filters, in both space and time.
Cuckoo filters are based on cuckoo hashing. A Cuckoo filter is essentially a cuckoo hash table which stores each key's fingerprint. As Cuckoo hash tables are highly compact, a cuckoo filter often requires less space than conventional Bloom filters for applications that require low false positive rates (< 3%).
This library was designed to provide a target false positive probability of ~P(0.001) and was hard-coded to use sixteen bits per item and four nests per bucket. As it uses two hashes, it's a (2, 4)-cuckoo filter.
A Cuckoo filter supports following operations:
cuckoo_filter_new(filter, max_key_count, max_kick_attempts, seed): creates a filtercuckoo_filter_free(filter): destroys a filtercuckoo_filter_add(filter, item, item_length_in_bytes): add an item to the filtercuckoo_filter_remove(filter, item, item_length_in_bytes): remove an item from the filtercuckoo_filter_contains(filter, item, item_length_in_bytes): test for approximate membership of an item in the filterexample/example.c: an example demonstrating the use of the filterinclude/cuckoo_filter.h: the public header filesrc/cuckoo_filter.c: a C-based implementation of a (2, 4)-cuckoo filtertests/test.c: unit testsTo build this example:
$ make testJonah H. Harris jonah.harris@gmail.com
The MIT License