The Research On Accurate Identification Method Of Elephant Flow In High-speed Network

High-speed network traffic measurement is the basis of real-time and accurate monitoring,management and control of the network at present,and elephant flow identification is an extremely critical task in network measurement,and it has a wide range of applications in the fields of network billing,congestion control,and traffic engineering.In view of the limitations of the existing hardware equipment,it is difficult to recognize the elephant flow in the highspeed network in consideration of storage capacity and processing speed.Therefore,the research on sampling methods and streaming methods applied to elephant flow identification is particularly important.However,neither of these two methods can achieve a better balance between resource expenditure and accuracy.This paper focuses on the sampling method and streaming method,combines the advantages of the two methods,in order to balance the resource cost and accuracy of elephant flow identification.The research work of this paper is as follows:(1)Analyze the principle of elephant flow identification in network measurement,sort out the sampling and data flow methods used for elephant flow identification,summarize common problems in elephant flow identification methods,and provide ideas for improving measurement accuracy and reducing resource overhead.(2)Aiming at the problem of low space utilization,an elephant flow identification framework based on filtering is proposed:Mice-Filtering Sketch(MF Sketch).According to the heavy tail distribution characteristics of network traffic,the structure uses the filtering strategy to differentiate the elephant flow and the mouse flow,which improves the space utilization and reduces the probability of the conflict between the elephant flow and the mouse flow.Then,the idea of "catching the large and discarding the small " is adopted to preserve the elephant flow as much as possible when the conflict occurs,so as to improve the accuracy of the elephant flow identification.The results of simulation experiments show that,when the same memory size is allocated,the proposed storage differentiation architecture has a significant reduction in error compared with the traditional Sketch-based data flow method.(3)In order to solve the problem of low processing performance in the filter,a shunt identification framework based on cuckoo hashing and sampling is proposed in this study:Shunt-Sketch.The structure uses the unique filtering mechanism of cuckoo hashing("kick out"operation)to deal with hash conflicts by using a deformable cuckoo filter to reduce the number of mouse flows being misfiltered.Then,the packets are shunted by the sampling method to reduce the processing burden in the filter and bypass the redundant processing operation of the elephant stream in the filter.The experimental results show that the shunt identification framework proposed in this paper can further improve the recognition accuracy on the basis of distinguishing storage architectures.