Fast Bloom Filters and Their Generalization

Bloom filters have been extensively applied in many network functions. Their performance is judged by three criteria: query overhead, space requirement, and false positive ratio. Due to wide applicability, any improvement to the performance of Bloom filters can potentially have a broad impact in many areas of networking research. In this paper, we study Bloom-1, a data structure that performs membership check in one memory access, which compares favorably with the k memory accesses of a standard Bloom filter. We also generalize Bloom-1 to Bloom-g and Bloom-Q, allowing performance tradeoff between membership query overhead and false positive ratio. We thoroughly examine the variants in this family of filters, and show that they can be configured to outperform the Bloom filters with a smaller number of memory accesses, a smaller or equal number of hash bits, and a smaller or comparable false positive ratio in practical scenarios. We also perform experiments based on a real traffic trace to support our filter design.