Efficient distributed adaptive transcoding: intelligent dynamic partitioning and virtual reference substream co-design

This paper proposes an intelligent dynamic partitioning algorithm (IDPA) integrated with virtual reference substreams (VRS) to optimize distributed transcoding systems. Traditional static and fixed-ratio partitioning methods suffer from inefficient resource utilization and poor load balancing in dynamic environments. IDPA dynamically adjusts computational load distribution using real-time cluster feedback, while VRS reduces overhead by simulating transcoding via metadata and lightweight frames. Experiments demonstrate a 38% reduction in transcoding time, 15% lower computational overhead, and 8% node load deviation compared to baseline methods. The solution enhances scalability and synchronization accuracy (timestamp offset ≤1ms). Future work includes extending protocol compatibility and integrating federated learning for adaptive model optimization.