100 Gbps Hash-Based Reconfigurable Pattern Matching

Paper in proceeding, 2026

High-Throughput pattern matching is crucial to applications such as network intrusion detection systems (NIDS). However, a pattern-matching engine typically cannot exceed a few Gbps unless it processes the input stream at a large stride, meaning multiple bytes per cycle. This requires searching for each pattern at multiple starting offsets in parallel, which increases hardware cost proportionally. This paper presents a new pattern-matching design that addresses this scalability challenge by efficiently scanning an input stream with a large stride while searching thousands of patterns. The proposed solution partitions the matching problem using a grouping heuristic, then employs a hash-based matcher for each partition. Each matcher searches a mutually exclusive set of patterns at specific offsets and produces at most one match at a time. To minimize the number of groups, our technique exploits the fact that most pattern-offset pairs (POPs) are mutually exclusive, enabling larger groups spanning multiple offsets, greatly reducing the resource cost. For a stride of N=64, our approach yields about 1/4 fewer groups and an order of magnitude less pattern memory compared to the conventional method of replicating resources N times. This enables a single FPGA, for the first time to our knowledge, to achieve 100 Gbps guaranteed throughput regardless of input data, while matching the entire SNORT NIDS ruleset.