Digital Twin-Driven Network Architecture for Video Streaming
Xinyu Huang, Haojun Yang, Shisheng Hu, Xuemin Shen
TL;DR
The paper addresses the challenge of managing next-generation video streaming services with highly variable demands by introducing a digital twin–driven network architecture (DTN4VS/VDTNet). It proposes three DT types (UDT, IDT, SDT) and a three-domain framework (physical, slice, DT) to enable real-time emulation, domain-separated control, and tailored resource management via network slicing and advanced AI. Key contributions include a data-importance based abstraction mechanism, a holistic DT performance metric, and a distributed transfer-learning approach for adaptive DT model updates, illustrated by a case study on DT-assisted network slicing for short video streaming. The work outlines open issues in DT coordination, closed-loop management, and security/privacy, highlighting practical pathways to deploy DT-based video streaming at scale and improve QoS/QOE while reducing computation overhead.
Abstract
Digital twin (DT) is revolutionizing the emerging video streaming services through tailored network management. By integrating diverse advanced communication technologies, DTs are promised to construct a holistic virtualized network for better network management performance. To this end, we develop a DT-driven network architecture for video streaming (DTN4VS) to enable network virtualization and tailored network management. With the architecture, various types of DTs can characterize physical entities' status, separate the network management functions from the network controller, and empower the functions with emulated data and tailored strategies. To further enhance network management performance, three potential approaches are proposed, i.e., domain data exploitation, performance evaluation, and adaptive DT model update. We present a case study pertaining to DT-assisted network slicing for short video streaming, followed by some open research issues for DTN4VS.