Research On Multi-source And Multi-path Congestion Control For NDN

In recent years,with the continuous development of infor-mation technology,many problems have gradually erupted in the TCP/IP architecture in the face of high-bandwidth requirements mainly for high-definition video applications.Named Data Networking(NDN),as a content-centric network architecture,can well adapt to the existing high bandwidth requirements.But the change of network architecture also brings many new challenges to the research of its congestion control mechanism.The congestion control mechanism affects the transmission efficiency of the network,so the research on the NDN congestion control mechanism is of great significance.As an important feature of NDN architecture,multi-source multi-path can improve the utilization of network resources by using it properly.By analyzing the characteristics of NDN multi-source multi-path,this thesis proposes a cache-aware Multi-Source and multi-Path Congestion Control mechanism(MSPCC)and a Reinforcement Learning based NDN multi-source and multi-path Congestion Control(RLCC).The main research work is as follows:1)In view of the new challenge brought by NDN multi-source multi-path to congestion control mechanism,multi-source multi-path identifica-tion mechanism and nearest data source search algorithm are designed in this thesis.On this basis,sub-path classification algorithm and multi-source multi-path congestion window adjustment scheme are also de-signed.The multi-source multi-path identification mechanism and the nearest data source search algorithm can be used to identify the differ-ent sub-paths of all data sources and deal with the cache miss problems existing in the identification mechanism.The sub-path classification algorithm and the multi-source multi-path congestion window adjust-ment scheme can divide the sub-path into congested and uncongested and adjust the congestion window according to the congestion status of each strip path.2)Aiming at the problem that the current network measurement index can-not reflect the network performance and the multi-path congestion con-trol rule is relatively simple,this thesis designs the multi-dimensional network state index.On this basis,this thesis designs a network con-gestion degree prediction model and a more fine-grained multi-source multi-path congestion window adjustment scheme.The network state index mainly includes the basic state information of the network,the basic information of the sub-path and the influence weight of the lost path.The network congestion degree prediction model can be trained asynchronously and the reinforcement learning model can be used to generate congestion degree prediction rules.In the multi-source multi-path congestion window adjustment scheme,the window adjustment mechanism can be used according to the congestion degree of the net-work.The experimental results show that MSPCC and RLCC can better adapt to the characteristics of NDN multi-source multi-path and improve network utilization compared with the other two comparison schemes(PCON and MPCC).With caching,MSPCC improves average throughput by about20% and 50% compared to MPCC and PCON,respectively,while RLCC improves throughput by about 20% compared to MSPCC.Without caching,MSPCC performs similarly to MPCC,improving average throughput by about 15% relative to PCON and average throughput by about 30% rela-tive to MSPCC.