Chucky: A Succinct Cuckoo Filter for LSM-Tree

Abstract

Modern key-value stores typically rely on an LSM-tree in storage (SSD) to handle writes and Bloom filters in memory (DRAM) to optimize reads. With ongoing advances in SSD technology shrinking the performance gap between storage and memory devices, the Bloom filters are now emerging as a performance bottleneck. We propose Chucky, a new design that replaces the multiple Bloom filters by a single Cuckoo filter that maps each data entry to an auxiliary address of its location within the LSM-tree. We show that while such a design entails fewer memory accesses than with Bloom filters, its false positive rate off the bat is higher. The reason is that the auxiliary addresses occupy bits that would otherwise be used as parts of the Cuckoo filter's fingerprints. To address this, we harness techniques from information theory to succinctly encode the auxiliary addresses so that the fingerprints can stay large. As a result, Chucky achieves the best of both worlds: a modest access cost and a low false positive rate at the same time.