Abnormal Traffic Detection And Load Balancing In Data Center Network

With the rapid development of emerging technologies and industries such as cloud computing、big data、artificial intelligence、and 5G,the growth of data center traffic around the world continues to accelerate.Many applications of network monitoring and management require to measure the entropy of network traffic.Traditional methods collect network traffic characteristics from switches and send them to the SDN controller to measure entropy which requires a trade-off between detection delay and overhead.The layer-4 load balancing system plays a critical role as an important network function for balancing traffic,traditional dedicated hardware load balancing systems and software load balancing systems are difficult to meet the requirements of low latency、high throughput、low cost、and flexible configuration.Current applications bring huge challenges to network abnormal traffic detection and load balancing in data centers:one is the challenge of implementing the above-mentioned scheme and system on the programmable data plane;the other is the challenge of improving the performance of the above-mentioned scheme and system.The data plane programming language P4 is reconfigurable,protocol independent,and platform independent,this paper designs an abnormal traffic detection system and a load balancing system based on P4,the specific research results are as follows:(1)An abnormal traffic detection scheme based on programmable data plane.Two algorithms,ESTLog and ESTDiv,are designed to perform logarithmic and division operations in the data plane,and they strike a balance between accuracy,processing latency,and generality.In addition,based on the above two algorithms and Count Sketch,a scheme for measuring network traffic entropy in the data plane-ESTEntropy is proposed,and abnormal traffic detection is performed according to the network traffic entropy.Simulation experiments show that the scheme has high accuracy and can reduce the processing delay by 86.3%-87.9%.(2)A memory-efficient layer-4 load balancing system.We design a layer-4 load balancing system Strainer based on the BBAS data structure to manage the connection state.This scheme is suitable for large-scale data centers,it only needs to manage the state of partial connections,which can save memory,reduce the burden of the controller,and effectively resist SYN flood attacks.The effectiveness of the system is verified by simulation experiments and experiments on hardware switch.