Cuckoo++ hash tables

Abstract

Hash tables are essential data-structures for networking applications (e.g., connection tracking, firewalls, network address translators). Among these, cuckoo hash tables provide excellent performance by processing lookups with very few memory accesses (2 to 3 per lookup). Yet, they remain memory bound and each memory access impacts performance. In this paper, we propose algorithmic improvements to cuckoo hash tables to eliminate unnecessary memory accesses, without altering the properties of the original cuckoo hash table so that all existing theoretical analysis remain applicable. We also present an implementation tailored to run efficiently on Intel Xeon processors, thus supporting NFV and softwarization trends and compare it to the optimized implementation of DPDK. On a single core, our implementation achieves 37M positive lookups per second (i.e., when the key looked up is present in the table), and 60M negative lookups per second, a 45% to 70% improvement over DPDK.