Bandwidth Allocation System Designed And Implementation For Knowledge Defined Network

With the increasing complexity of the network scale,the types of services are different and have significant differences in bandwidth resources Qo S for the network.The traditional network architecture cannot achieve flexible bandwidth control due to its relatively independent distributed management and control mechanism.It cannot meet the dynamic changes of multi-service requirements and the adaptiveness of the management and control system.Knowledge Define Network is a new architecture that combines the knowledge plane with the Software Defined Network architecture.The knowledge plane provides a computing carrier for machine learning models,then SDN provides numerical control separation.The network management helps the network to manage bandwidth resources more flexibly.It is important to provide flexible and adaptive bandwidth allocation schemes for multiple services,when the services are faced with the dynamically changing bandwidth requirements in the network.Under the KDN architecture and realize the maximum utility value of the network is an important research topic.Traditional bandwidth allocation schemes measure network operation efficiency from performance indicators,such as delay throughput and jitter.And those schemes evaluate the available benefits of the network from the dimension of utility value.However,the single optimization goal is difficult to achieve a balance between performance and utility.It is difficult to meet the Qo S of different services.Aiming at the multi-service scenario with different Qo S,this thesis proposes a utility bandwidth allocation scheme,which based on DQN to achieve dynamic bandwidth management of multi-service requirements.For different allocation targets,the corresponding utility value is quantified.The DQN model interactives the network state,which is used for training to adapt to the dynamically changing bandwidth requirements and obtain the allocation result of the maximum utility value.The result shows that this scheme can get the maximum utility allocation bandwidth,and adapt to the dynamic change of bandwidth demand to improve the link bandwidth utilization rate.It has advantages in both network utility and performance.In order to improve the flexibility and adaptability for management and control of the network system,this thesis studies the characteristics of the knowledge plane and the KDN architecture in detail.Designs and implements a knowledge plane prototype that based on the DQN model.The KDN network architecture is built by integrating the knowledge plane,control plane and data plane.Then design and implement the bandwidth system and allocation method for KDN.The system includes four functional modules,such as traffic collection,traffic identification,routing planning,and bandwidth allocation.These modules work together to achieve the unified and effective management,which include the traffic,routing,and bandwidth resources for the KDN architecture.The validity of the system platform and module functions are verified by building system test.The result shows that the system can achieve effective bandwidth allocation,and improves the operation efficiency of the management system with multiple service requirements.