Research On High Performance Network Traffic Diversion Technology

As the network bandwidth continuously increases,the processing efficiency of traditional standalone content audit system can no longer meet the requirements of large-scale network traffic.Therefore,the parallel processing technology has been widely applied and the traffic diversion technology has become a popular research subject.To reduce pressure on the processing node,the traffic diversion technology for the network service is introduced.This essay studies and optimizes this technology from the perspectives of service identification and load balancing to improve its performance.Firstly,this essay studies on the service identification method.Currently,the application-layer load identification has high accuracy.However,services are filtered packet by packet,which may lead to performance deterioration.The service-level identification caches the triplet on servers to effectively accelerate identification,but it lacks efficient caching algorithms.This essay proposes a service traffic DPI model based on LRFU caching algorithm status records.The service-level identification uses the LRFU-QT caching algorithm that considers both time and frequency.It stores low-frequency nodes in chain queues to prevent frequent swapping in and out and conducts the periodic cleaning on nodes using the time wheel.At last,a comparison experiment is performed to verify that the LRFU-QT has a higher cache hit ratio and processing efficiency,and can effectively improve service identification efficiency.Secondly,this essay studies on the load balancing model.Nowadays,the load balancing model that combines the static HASH and dynamic feedback methods can provide a good load balancing effect,but sessions may be broken.To solve this problem,this essay proposes the CSAF model based on the predicted network traffic load balancing model.This model uses the dynamic HASH method for traffic allocation balance during basic distribution and adjustment,and implements zero session breaking using the flow table and dynamic exponential smoothing method.An experiment is then performed to prove that,this model can achieve zero session breaking with load balance and low packet loss rate.Finally,this essay designs and implements a high-performance network service traffic diversion system based on the preceding models,gives an implementation diagram and function description of the system,and performs the function and performance tests on the system.The test proves that the network service traffic diversion system can provide a high throughput and is valuable in engineering applications.