Abstract
Industrial Internet of Things (IIoT) generates a massive of mixed traffic, which shares the same bottlenecked network resource. The burst of flows having heavy-tailed property imposes a challenge in provisioning stringent quality of services (QoS) to light-tailed flows and fair QoS among flows. Furthermore, the flows following the same tail distribution also require distinct QoSs. For example, video streams for closed-loop control and for environment monitor tolerate distinct delays. The complicated traffic characteristics and QoS requirements demand a flexible QoS scheme. We adopt a weighted remaining lifetime to model the emergency of each packet, based on which we propose a delay-based scheduling framework to provide packet-granular services. The framework adopts light-queue and heavy-queue to buffer the corresponding flows. Then, dynamic weight earliest deadline first (DWEDF) is proposed to position the arrival packets in the queue, and reinforcement learning is used to depart head of line packets of queues. The parallel-running of intra-queue buffering and inter-queue scheduling leads to a scalable QoS provisioning with low complexity. Simulation results verify the efficiency of our proposal in providing hard delay guarantee for real-time IIoT applications having light-tailed property and low delay bound violation ratio, delay-based fairness to heavy-tailed flows tolerating soft delay bound.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call