Analysis And Research Of Packet Matching Based On Intelligent Algorithm

Network devices such as firewall and router, have the basic function of the recognition and match of packets. To provide different services is a new requirement which network operators are facing. Data packets, which through network or host, must be monitored and matched with the rules, which are in the rule library, by these network devices. Many scholars put forward the corresponding packet matching algorithm, which have their advantages and disadvantages. Algorithm, which is based on the software, can't satisfy the requirement of the increasing link speed, and algorithm based on the hardware, can't meet the growing scale of network.Intelligent algorithm, which is simple, has less control parameters and better performance, has caught the interests of many researchers. Their superior ability to solve the optimization problem provides a new idea of thought to solve the problem of packet matching. This article aims to systematically research for intelligent algorithm to solve packet matching, and focuses on how to solve the large-scale multidimensional packet matching problem by differential evolution algorithm. On the subject, this dissertation carried out the research work, which mainly includes:(1) This dissertation reviewed a lot of literature about research status of packet matching algorithm, and researched the classification of packet matching algorithm, and did detailed analysis of the pros and cons of each packet matching algorithm. Based on the existing study, this dissertation carefully researched the performance evaluation indexes of packet matching algorithm.(2) The packet matching problem was understood as a function mapping, which map high-dimensional rule space to a low dimensional space. The dissertation integrated the differential evolution algorithm which is based on real number encoding, and the traditional packet matching algorithm. The variation coefficient of thought has been applied in the adaptive value design. Through analyzing the characteristic of the distribution of mapping space, algorithm adaptive adjusted intensity of variation. Because the time performance of packet matching of proposed algorithm has a relatively weak correlation with the number of rule in the rule library, it is suitable for processing large packet matching problems. Due to the reason that differential evolution was applied to solve a practical problem, the data set is relatively stable, and the data has some common characteristics, Therefore, this article wanted to provide selection basis of mutation operator DE/X/Y/Z and the control parameter setting and the choice of the scheme of the evolution operation, by numerical experiments. Through numerical experiment to contrast the pros and cons between the proposed algorithm and traditional algorithm, experiment certificated that the proposed algorithm effectively improved comprehensive performance in speed and storage space. By studying, the proposed algorithm aimed to explore the general ideas and methods to solve practical problem by differential evolutionary algorithm.(3) This dissertation analyzed packet matching problem from the angle of control theory, and put forward a new matching algorithm, which combined evolutionary algorithm and neural network to solve packet matching. First used evolution algorithm to implement the dimension reduction of characteristics of the data set, and used differential evolution algorithm to set weights and select activation function. Then simplified neurons of hidden layer by impact factor. Finally, fine-tuned neural weights using back propagation algorithm. Numerical experiments showed that this algorithm can effectively improve packet matching comprehensive performance, and it had much less correlation between the number of rules and packet matching performance, and compared with traditional algorithm, it can effectively solve large-scale rule library matching problem. The proposed algorithm was aimed to research the general of thought of combination between differential evolution algorithm and other intelligent.(4) This dissertation took packet matching problem as the search problem in high dimension space and described packet multi-layer characteristics for packet matching by creatively using multi-layer foundation base. In each layer, applied double populations differential evolution algorithm to extract bit base and entity base, respectively, and used the average self information and average mutual information to measure the fair or fault for foundation base. The proposed algorithm was able to select the number of bit base and entry base, according to actual scale of rule library. This method adapts well to the rule number growth of the rule library. From the results of numerical experiment, the proposed algorithm achieved very good effect.(5) According to the sample information of population currently, this dissertation introduced information entropy and histogram into information statistics of population currently, and used these statistics information to adjust dynamically relevant parameters of optimization algorithm of Chemical-reaction-inspired meta-heuristic. This dissertation, for the first time, analyzed problem from the sample of population currently, instead of assuming what is the distribution of all sample. From the experimental results, the algorithm proposed by this dissertation achieved satisfying expected effects. For dynamic adjustment parameters of intelligence algorithm of Chemical-reaction-inspired meta-heuristic, the relation of scale and performance of packet matching is more loose. And the intelligent algorithm proposed is more suitable for packet matching.From the characteristics of the data set itself, this article analyzed and researched how to solve the large-scale multidimensional data packet matching problem by intelligent algorithm. Differential evolution algorithm was mainly studied, and its fusion with neural networks was too researched. Moreover it integrated layered thought with differential evolution algorithm to reduce the dimension and scale of the problem. Algorithm of Chemical-reaction-inspired meta-heuristic is a new type of genetic algorithm. Its capacity to process packet matching was analyzed by comparing it with the difference evolution algorithm to deal with the performance of packet matching.