Research And Design Of Performance Optimization Scheme For Pub/Sub System In SDN Network

A topic based Pub/Sub system is classified as a distributed message transmission middleware,which classifies messages according to their characteristics,and realizes the decoupling between message sender(publisher)and message receiver(subscriber).The SDN network has the characteristics of centralized control,and realizes the separation of control and forwarding.The port of SDN switch supports priority queue with different bandwidth.The bandwidth of queue directly determines its ability to forward data.The combination of the Pub/Sub system and the SDN network can transmit messages of different topics with different queues to achieve thematic differentiated services.At present,the combination of Pub/Sub system and SDN is not perfect,and there is still a lot of space in the adjustment algorithm of queue bandwidth.The adjustment algorithm of queue bandwidth is to adjust the queue bandwidth dynamically according to queue delay and priority,so as to adapt to the real time change of the network.However,the existing method of obtaining queue delay is not very accurate,which results in the poor effect of bandwidth adjustment.At the same time,the adjustment of queue bandwidth is limited to a single network node,which does not take into account the user's need for end to end delay.In addition,the system needs to be improved and optimized in terms of traffic management and interface performance.Based on the current status of research,this paper puts forward the optimization scheme of the performance of the system:one is to put forward two methods to obtain the queue delay.One is incremental difference method,the number of bytes of incremental import queue cleverly avoid the influence of time difference on the queue length,while reducing the error caused by packet loss by another;Another is the XGBoost model,XGBoost model is used in machine learning data queue of training model,and then enter the data prediction queue current queuing delay.Secondly,we propose a queue bandwidth adjustment algorithm based on user needs.According to users' demand for end to end delay,we adjust the delay constraint value of each node,so that the total delay of the entire path is no more than that of users.The third is the optimization of system traffic management and interface performance.By proposing a series of control strategies,we can achieve traffic information backup,interface attack prevention,change the underlying structure of the interface,improve the efficiency of its call,and achieve high availability of the system.The original Pub/Sub system has achieved good performance after the optimization of the above schemes.It has been optimized in terms of delay,packet loss rate,throughput,stability and the efficiency of the interface call is increased by 20 times.At the same time,the system is deployed in the network environment built by entity SDN switches for the first time.After a lot of tests and verification,the validity and availability of the above schemes are confirmed.