.png){kind=icon}
.png){kind=icon}
Return
Optimizing Time-Sensitive Software-Defined Wireless Networks With Reinforcement Learning
.png){kind=icon}
OA
AI
DOI：10.1109/ACCESS.2022.3222060.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Even though wireless networks are inevitable in mobile or infrastructure-less communication systems, such as vehicle-to-everything (V2X) infrastructure in automobile, precise formation control of unmanned vehicles (UVs), or other industries that employ ad hoc deployment of systems, operation and maintenance of network applications additionally impose time constraints on the wireless network. Such the requirement poses an immediate challenge to the time-sensitive aspects of devices, applications and network control, which has been addressed in the realm of time-sensitive networking (TSN). Meanwhile, software-defined networking (SDN) has successfully presented its efficiencies in ensuring quality of service for network traffic to accommodate many functions of network control and management. In this regard, we propose a traffic engineering solution based on reinforcement learning (RL) to implement TSN links with SDN over a wireless network, then optimize the quality of TSN links, and protect background traffic from excessive resource allocation for TSN-enabled but SDN-supported traffic. We implemented SDN-based TSN on a real testbed, consisting of real nodes as single board computers (SBCs) and an SDN controller, and applied RL-based network control solution to the network. The empirical results are promising in that the jitter of time-constrained traffic is improved by 24.6% and throughput of background traffic is increased by 6.5%, compared to the manual configuration mode.
Keywords:
Reinforcement learning
time-sensitive network
resource allocation
traffic control