A Dynamic Resource Allocation based on Network Traffic Prediction for Sliced Passive Optical Network
Xuanqiao Liang, Qinghua Tian, Fu Wang, Wensheng Yu, Xiangjun Xin
Abstract
As the rapid development of the Industries Internet of Things (IIoT), the access network characters with massive connections, deterministic latency and large bandwidth. Therefore, the existing time-division-multiplexing Passive Optical Network (TDM-PON) technology needs to evolve from the original bandwidth scheduling pattern into the cooperative scheduling with multi-network slices (NS). Meanwhile, supporting IIoT and Mobile Fronthaul (MFH) in the TDM-PON via fixed network slicing is challenging, as MFH services may generate microburst traffic, thus impacting the performance of IIoT applications. In view of this, we propose a resource allocation scheme using LSTM neural network to predict traffic in an SDN-based TDM-PON. Our approach considers a network slicing architecture that meets the diverse requirement for both MFH and IIoT services. The results show that our algorithm can reduce the MFH slice latency by 23.3% compared with the typical method (load of 0.8).