Cuckoo Filter: Simplification and Analysis

TL;DR

This paper introduces a simplified cuckoo filter with fewer hash calls and provides the first theoretical guarantees on its performance, demonstrating high success probability with sufficiently large fingerprints.

Contribution

It simplifies the cuckoo filter design and establishes the first theoretical performance guarantees for its success probability.

Findings

Simplified cuckoo filter with fewer hash function calls.

Theoretical guarantees on success probability.

High success probability with large fingerprint sizes.

Abstract

The cuckoo filter data structure of Fan, Andersen, Kaminsky, and Mitzenmacher (CoNEXT 2014) performs the same approximate set operations as a Bloom filter in less memory, with better locality of reference, and adds the ability to delete elements as well as to insert them. However, until now it has lacked theoretical guarantees on its performance. We describe a simplified version of the cuckoo filter using fewer hash function calls per query. With this simplification, we provide the first theoretical performance guarantees on cuckoo filters, showing that they succeed with high probability whenever their fingerprint length is large enough.