Resourch Allocation For Radio-Over-Fiber Access Network In Cloud Computing Environment

As an essential infrastructure supporting future mobile communications,the radio-over-fiber access network can provide high-speed and flexible user access service by combining the optical network's high-speed and large-capacity with the wireless network's flexible mobility.In the development process of the radio-over-fiber access network,there are a wide variety of transmission equipment,diverse supporting services and massive number of nodes.This has resulted in the limited scale expansion and low resource utilization faced by the radio-over-fiber access network.The challenge of efficiency has forced the radio-over-fiber access network to expand from a traditional closed system to an open cloud computing environment.In recent years,the cloud-based radio-over-fiber access network has become a new generation of key technology that meets the needs of users' access network by virtue of its advantages of high flexibility,large bandwidth,and high reliability.In the context of the rapid development of cloud computing,the number of devices,deployment scales,and various communication protocol standards has further increased.The radio-over-fiber access network in the cloud computing environment has shown heterogeneous characteristics of multiple scenarios and multiple technologies and faces resources.There are many difficulties,such as restricted distribution,complicated traffic scheduling,and weak failure recovery.Specifically,there are three main application scenarios for the resource allocation of the radio-over-fiber access network based on cloud computing.From the inter-data center network perspective,long-term network resource planning can avoid global resource waste and improve service scheduling efficiency.From the intra-data center network perspective,accurate burst traffic scheduling can avoid burst traffic congestion and performance degradation and improve the cloud services' response speed.From the mobile fronthaul network perspective,across-scenarios resource allocation and rapid and accurate failure positioning can improve the network's overall performance and reduce economic losses due to waste of resources and failures.Key issues such as long-term resource planning for networks between data centers,burst traffic scheduling in data centers,network slicing,and rapid failure recovery of mobile fronthaul networks have emerged around different scenarios.In response to the problems mentioned above and challenges,this paper focuses on the radio-over-fiber access network's resource allocation in the cloud computing environment.It uses artificial intelligence technology to assist the three applications:inter-data center network,intra-data center network,and mobile fronthaul network.Select the appropriate deep learning method for the scenario,optimize and redesign the neural network model,use the learning results such as traffic prediction and alarm classification to guide the subsequent resource allocation process.And finally,the performance verification of the radio-over-fiber access network in the cloud computing environment was carried out on the simulation platform.Relevant research work has essential reference significance for the broad application of radio-over-fiber access networks and efficient resource allocation in the cloud computing environment.The main innovations are listed as following:(1)In the inter-data center networks,for the problem of long-term network resource allocation limitation.We first introduce traffic prediction,called Multiple Time Interval Feature-Learning Network(MTIFLN),to assist the traffic scheduling in the radio-over-fiber access network.Secondly,a resource allocation plan based on traffic prediction is designed.The efficient allocation of network resources among data centers is realized according to traffic prediction and current network resource utilization.By integrating multiple bidirectional recurrent neural networks with different sampling intervals into a framework,the MTIFLN model has a powerful ability to extract flow characteristics at different time intervals.Thus,avoid the cumulative error of multi-step predictions,thereby achieving one-step accuracy of long-term flow prediction.In the resource allocation scheme based on traffic prediction,the traffic priority is calculated based on the prediction result,and existing resources and resources are reserved for future traffic.The simulation results show that the MTIFLN model can effectively improve the accuracy of long-term traffic prediction.The resource allocation scheme based on traffic prediction can effectively utilize the radio-over-fiber access network's network resources.(2)In the intra-data center networks,we proposed an error feedback-based Spiking Neural Network(SNN)model to achieve high-precision prediction of burst traffic for the complicated traffic scheduling problem.Then,we developed a predictive auxiliary scheduling algorithm to perform the worst-case efficient scheduling of burst traffic.On the one hand,the SNN framework based on error feedback can significantly enhance the ability to extract burst traffic features by imitating the biological neuron system.The multi-synaptic mechanism can effectively extract the burst flow characteristics,and the error feedback module can realize the back propagation of the error without demand guidance.On the other hand,the forecast-assisted scheduling algorithm uses global evaluation factors and traffic scaling factors to arrange well-predicted traffic.Simulation results show that the proposed method can effectively integrate SNN into the traffic scheduling scheme and achieve satisfactory performance with affordable computational complexity.(3)In the mobile fronthaul network,the optical network's resources and the wireless network are difficult to achieve global allocation because of the problem that a data-driven method is first proposed.By mapping the resource state to a two-dimensional vector diagram,the network from the optical network to the wireless network is realized.Combined with network slicing,this design can significantly shorten the time for slice status updates.The corresponding slice reconfiguration algorithm is introduced to balance each slice's load in the optical network and the wireless network.The vector diagram comprehensively considers the resource relevance and load information of the mobile fronthaul network and the data center network.Simulation results prove that the proposed method can significantly reduce slice state update costs and obtain ideal slice reconfiguration performance.(4)In the mobile fronthaul network,in response to the inefficiency of fault recovery,an innovative Deep Belief Network-based Fault Location(DBN-FL)model is proposed to achieve a single chain of mobile fronthaul networks.Fast location of road faults.The DBN-FL model consists of two stages,including a hybrid pre-training stage and a fine-tuning stage based on the Levenberg Marquardt(LM)algorithm.We will combine supervised and unsupervised learning in the mixed pre-training stage to reduce training samples' demand.In the fine-tuning stage,the LM algorithm is used to replace the traditional back-propagation algorithm to fine-tune the DBN-FL model,which greatly reduces the time required for offline training.Experimental results show that the proposed DBN-FL model can achieve high-precision fault location as a classifier(with an accuracy of more than 96%).It is superior to traditional deep learning methods in terms of location accuracy and training efficiency.