Spatial Encoding of Flow Spaces for Intelligent SDN Applications

Abstract

Efficient encoding of network flow spaces while preserving spatial locality is essential for intelligent Software-Defined Networking (SDN) applications, particularly those employing reinforcement learning (RL) methods in a reactive manner. In this work, we introduce a spatially aware Bloom Filter-based approach to encode IP flow pairs, leveraging their inherent geographical locality. Through controlled experiments using IoT traffic data, we demonstrate that Bloom Filters effectively preserve spatial relationships among flows. Our findings show that Bloom Filters degrade gracefully, maintaining predictable spatial correlations critical for RL state representation. We integrate this encoding into a DQN-based eviction strategy for reactive SDN forwarding. Experiments show that Bloom Filter-encoded, spatially aware flow representation enables up to 7% and 8% reduction in normalized miss rate over LRU and LFU, respectively, across 10 hours of traffic, demonstrating potential for low-latency applications. This experiment justifies the usefulness of preserving spatial correlation by encoding the flow space into a manageable size, opening a novel research direction for RL-based SDN applications.