A Genetic Programming approach for hardware-oriented hash functions for network security applications

Non-cryptographic (NC) hash functions are generally designed for speed and efficiency, which enables their use in many network security applications that require efficient lookup and counting, such as Bloom Filters and Count-Min (CM) Sketch structures. The performance of these structures heavily relies on underlying hash functions. Therefore, any advancement in the hash function design significantly impacts the overall performance of these structures. This paper presents a novel family of 32-bit NC hash functions (NCGPH-32) evolved using Genetic Programming (GP) and their corresponding implementation on Field Programmable Gate Arrays (FPGAs). This family of NC hash functions generates smaller hash values concatenated to produce larger hash outputs. Inspired by related work on 96-bit NC hash functions with GP, this work optimizes the performance of 32-bit NC hash functions on FPGA while achieving high scores on specific avalanche metrics (avalanche dependence, avalanche weight, and entropy) when considering concatenated 96-bit outputs. This optimization is of utmost importance to address the escalating demand for Terabit Ethernet networks, specifically in processing real-time network flow IDs (identification and monitoring) at line rate. The throughput, latency, operating frequency, and resource utilization are evaluated on an FPGA and compared against 17 state-of-the-art NC hash functions. The results show that the proposed 96-bit concatenated hash function surpasses prior GP-based and other state-of-the-art NC hash functions by at least 36% in operating frequency, 30% in throughput and reduces latency by 27%. The demonstrated improvements in the hash design not only cater to the present demands of Terabit networks but also meet the expected near-future demands. Additionally, we integrate these hash functions into the Standard Bloom Filter (SBF) architecture and demonstrate comparable false positive rates (FPR) to state-of-the-art NC hash functions, affirming their effectiveness and applicability. We have also conducted several statistical tests on hash outputs of NCGPH-32 to demonstrate the high random nature and uniform distribution.