HeatCache: A Heat-Predictive TCAM Rule Caching Framework with Dependency-Aware Optimization

Ternary Content Addressable Memory (TCAM) enables fast parallel lookup in Software-Defined Networking (SDN) switches but is limited by its high cost and power consumption. Therefore, integrating TCAM and Random Access Memory (RAM) to form a caching system has become the mainstream solution. However, prior approaches suffer from two key issues: (i) lack of predictive analysis for identifying hot rules, leading to suboptimal cache-hit rates; (ii) inefficiency for handling rule dependencies, requiring extra TCAM resources or generating massive rule fragments. To address these challenges, we propose HeatCache, a TCAM-based rule caching framework that enables complementary rule heat prediction to filter hot rules more accurately via a cluster model and a self-exciting point process. For rule dependency management, HeatCache introduces a Heat-driven Segmentable Dependency DAG (HSD-DAG) data structure for two-stage flow tables, mitigating rule fragments and enhancing TCAM utilization. Experimental results show that HeatCache outperforms state-of-the-art schemes by up to 26.6% in terms of TCAM cache-hit rate.