Deep Reinforcement Learning Based Routing in IP Media Broadcast Networks: Feasibility and Performance

The Media Broadcast industry has evolved from Serial Digital Interface (SDI) based infrastructures to IP networks. While video broadcast is well established in the data plane, use of networks transport media flows still poses challenges terms resource management and orchestration. SDN orchestration architectures have emerged that route a service across provider network. Several approaches multimedia flow routing been proposed context streaming applications over Internet. These range model linear optimization solutions high complexity simple shortest path heuristics with either static or dynamic link costs. More recently model-free methods such as Deep Reinforcement Learning (DRL) for Traffic Engineering scenario however specific requirements, services like Master Control Room operation Live broadcasting events, it rarely addressed literature. In this work we propose DRL method compare cost algorithms on Dijkstra paths. This our knowledge first follow approach infrastructures. algorithm designed considering specifications capabilities one leading orchestrators market considers more common Service Level Agreement requirements industry. Three different are implemented compared evaluate them using real network topology. results indicate applicable production scenarios achieves considerable performance gains when commonly used today.