Research On Resource Management Optimization In Optical Data Center Networks Based On Machine Learning

With the development and maturity of advanced technologies such as artificial intelligence(AI),cloud computing(CC),big data(BD)and internet of things(IoT),software-defined network(SDN),machine learning(ML),deep learning(Deep Learning,DL)and other interdisciplinary technologies,the optical data center networks(ODCNs)structure is flattened and the networking mode is intelligent,resource virtualization and other directions are evolving rapidly.At present,due to the continuous expansion of the scale and capacity,the topology has become more complex and functions are more dynamic.The overall performances of the ODCNs depend on many factors.Various dynamic performance indicators are becoming more and more important,and service carrying requirements become more diverse.Traditional spectrum allocation algorithms are difficult to meet the resource requirements of different services,resulting in fragmentation of the spectrum.In addition,the point-to-point static single parameter monitoring technology is difficult to meet the analysis needs of large scale dynamic and complex optical networks,and the analysis of fault is not comprehensive,timely and accurate.In view of the shortcomings of existing methods and research,the specific research contents are as follows:(1)This paper first designs architecture of the software-defined optical network(SDON)that introduces artificial intelligence technology.And through the introduction of artificial intelligence technology at the control level to assist the SDON controller,it is realized flexible management of capacity,fault prediction,channel quality assessment,and traffic monitoring.The characteristics of traffic flows and architecture between inter-DC and intra-DC.(2)Aiming at the problem that the spectrum allocation algorithms cannot meet the needs of different services,this paper proposes an algorithm that machine learning based on the flexible resource management algorithm(Machine learning based on the flexible resource management,ML-FRM),which uses the support vector machine algorithm to classify the channel into 4 categories;the unsupervised learning K-Means algorithm is used to cluster the traffic flows;according to the result of the clustering results,different service levels are matched to different channels and the appropriate spectrum resource allocation algorithms are allocated.(3)Aiming at the problem of fault analysis in the optical data center networks,this paper proposes the Long Short-term Memory-Support Vector Machine(LSTM-SVM)algorithm,which uses the LSTM algorithm to track the states of the parameters,and then imports the predicted parameter values in the trained support vector machine model,the fault prediction of the optical data center networks is realized.The ML-FRM algorithm proposed in this paper makes full use of the powerful mining and reasoning capabilities of machine learning,and closely combines existing network data to achieve multi-objective parameter analysis in the optical data center networks;the proposed LSTM-SVM algorithm monitors the data in three-dimensional based on the integrated and deep learning algorithms,it completes the state tracking and prediction of parameters,and further realizes efficient,accurate and intelligent fault prediction.Also,it provides effective fault analysis models and comprehensive analysis solutions for dynamic,reliable and intelligent technology of optical data center networks.Therefore,the optical data center networks can respond to multi-service requirements,reduce costs,and achieve high resource utilization accurately and quickly.