The growth of multimedia applications such as live streaming, online gaming, and virtual conferencing has intensified the demand for robust and efficient data transmission over heterogeneous and dynamic access networks using a multipath streaming scheduler. Traditional model-based multipath schedulers lack adaptability. Although learning-based approaches promise improved adaptation, their centralized architectures introduce challenges in scalability, user privacy, and responsiveness. Furthermore, relying solely on learning at individual devices is often insufficient due to the limited volume and diversity of local data, making it challenging to derive robust and generalizable policies. To address these limitations, we propose Fed- MVP, a federated multi-homing video streaming protocol within a hierarchical cloud-edge architecture. Fed-MVP decentralizes learning and decision-making processes to edge data centers, enhancing scalability and preserving user privacy. It employs fine-tuned meta-models at the edge for real-time adaptation to local network conditions and utilizes Proximal Policy Optimization (PPO) for efficient learning. We introduce a Modified Federated Averaging (Mod-FedAvg) algorithm for effective aggregation of model updates from clients. Trace-driven emulations highlight Fed-MVP’s superior performance, demonstrating up to 61% and 43% improvement in convergence rate during test and training phases, respectively, up to 40% reduction in download time, up to 35% enhancement in video quality assessments, and up to 38% reduction in stalling time compared to state-of-the-art multipath schedulers.

Video streaming performance may degrade substantially in a mobile environment due to fast-changing wireless links. On the other hand, to provide ubiquitous services, heterogeneous static and mobile access and backbone networks will be integrated in the sixth-generation (6G) systems, so mobile users can take advantage of multiple access options for better services. Multi-path transport-layer protocols like Multi-Path QUIC (MPQUIC) show promise in utilizing multiple access links to address the impact of mobility. However, the optimal link selection that aims to provide statistical QoS guarantee for video streaming in a mobile environment with both user mobility and network mobility remains an open issue. In this paper, based on a lightweight Multi-Armed Bandit (MAB) technique, we develop aQoS-drivenContextualMAB(QC-MAB) framework for MPQUIC, which makes an intelligent access network selection and adaptively enables FEC coding to trade off delay, reliability and goodput. Extensive simulation results with ns-3 show that the proposed QC-MAB framework can outperform the state-of-the-art solutions. It achieves up to ten times lower video interruption ratio and three times higher goodput in highly dynamic mobile environments.

Multi-homing technologies are promising to support seamless handoff and non-interrupted transmissions. Scheduling packets across multiple paths, however, has the known issue of out-of-order (OFO) due to the heterogeneity of the paths, which is detrimental to users’ quality of experience (QoE). Wireless link characteristics undergo a fast change over time in mobile environments, thus aggravating the OFO issue. In this paper, we present a novel mobility-aware multipath QUIC (MMQUIC) framework in which interactions between link and transport layers are introduced so that the scheduler at a mobile sender is aware of uplink variations, and a new ACK packet structure is designed to inform the scheduler of downlink variations when the receiver is mobile. Based on MMQUIC, a Mobility-Aware Multipath Scheduler (MAMS) is developed, which forecasts the path conditions in successive time slots based on historical and current end-to-end (E2E) path conditions, along with wireless uplink/downlink conditions, and pre-allocates packets on multiple paths accordingly. We conduct a series of experiments to evaluate the performance of MAMS using network simulator 3 (ns-3). Simulation results demonstrate that MAMS effectively leverages the information related to mobility, achieving substantial performance gains w.r.t. the goodput and packet delay distribution under different mobility patterns.

The deployment of multipath transport protocols in the mobile environment can enhance the performance of delay-sensitive applications by enabling the simultaneous use of several network paths, resulting in faster data transmission. However, due to the heterogeneity of network paths, packets may not arrive on time or in order, affecting the performance of delay-sensitive applications. Therefore, a well-designed multipath scheduler is important to distribute data packets efficiently to guarantee the per-packet delay requirement. In this paper, we propose Meta-DAMS, a delay-aware learning-based multipath scheduler, aiming to ensure that end-to-end delay is below a predefined threshold for delay-sensitive applications. We introduce a hybrid meta reinforcement learning (meta-RL) architecture for Meta-DAMS in which offline meta-RL and online meta-RL are used to learn the optimal scheduling policy quickly and accurately in response to highly dynamic network conditions. Based on trace-driven emulation experiments, we demonstrate that Meta-DAMS surpasses state-of-the-art MP schedulers, ensuring a delay of 50 ms or less for 98% of packets after sufficient operational episodes, compared to the 83% achieved by existing MP schedulers. Even in initial operational episodes, Meta-DAMS maintains its superiority, guaranteeing 94% of packets with a delay of 50 ms or less, while the performance of the DQN-based MPQUIC scheduler drops to 72%. Meta-DAMS exhibits nearly triple the efficiency in terms of runtime compared to the DQN-based MPQUIC scheduler across varying episode numbers.