Konstantinos Christodoulopoulos

We introduce a dynamic Software-Defined Key Management System (SD-KMS) framework designed as a modular orchestration platform to overcome the limitations of static routing in Quantum Key Distribution (QKD) networks. The central Controller manages routing via three distinct mechanisms: an active polling interface for external solvers, a passive injection API for enforcing custom commands, and an embedded Dijkstra fallback engine to guarantee baseline service. Paired with this open architecture, we implemented a novel batch-based Linear Programming (LP) routing engine. This algorithm utilizes a "just-enough" proactive routing strategy that computes multi-path routes in polynomial time. By periodically processing demand batches, it optimizes end-to-end services to stay above safety thresholds, guaranteeing QoS by immediate key provisioning for dynamic application demands while allowing physical link pools to seamlessly absorb inherent QKD generation fluctuations. The capabilities of the SD-KMS and the LP engine were experimentally validated on a field-deployed, 5-node, multi-vendor QKD ring network. Functioning as a unified overlay across heterogeneous QKD systems, the framework successfully navigated a dynamic evolution scenario featuring key generation fluctuations, varying demands, and an intentional link failure. During periods of heavy key demands, the LP engine effectively performed load balancing by dynamically routing a portion of relay flows to secondary paths, preventing pool starvation and outperforming standard single-path solutions. When an active link was deliberately disrupted, the engine instantly redirected all traffic to the remaining functional path to ensure uninterrupted service continuity. Finally, upon link restoration, the system automatically engaged multipath routing across both available paths, maximizing the total key provisioning rate to optimize the volume of served request.