Research On Some Key Technologies Of Information Network Based On Theory Of Complex Networks

There has been some research progress in the field of complex network theory, but most of the achievements have not been applied in information networks and many related techniques should be optimized and improved before application. Based on previous researches in this filed, this dissertation studies in depth the problems such as admission control, QoS multicast routing, malicious software spreading, specific services identification in P2P networkBy researching complex network theory and applications, this dissertation has the contributions as below:A simulation platform for complex networks is built. According to requirement, a large scale ER model, BA model or NW model can be created in adjacent matrix or node information network model form, network statistic charateristics such as node's degress distribution, average path length and aggregted degress coefficient can be computed, an implementation and optimum algorithm is presented. This dissertation puts forward the node information network model method used for describing network, this method is of small buffer space and fexible strcture, overcoming the limit of computer processing capacity for using adjacent matrix representing network topology information, provides an efficient research platform for simulation of large scale complex networks and the spreading dynamics on complex networks as well as formation of discrete propagation dynamics model.Aiming at complex network, a new concept is put forward,it is node active degree that is used to express node's capability to attract other nodes and connect them, this concept playe an important role for deciding whether a node is connected to others, the higher node's active degree is, the larger connection probability is. Combined with the weight of node's active degree, an admission control algorithm based on complex network is presented. Different admission strategies are applied under the different bandwidth resource, when the occupied maximum bandwidth of all the services does not surpass predefined bandwidth threshold, by means of self-adaptable buffer range parameter adjustment, a reasonable limit to admission condition is implemented. After corresponding adaptable processing, the system can reach a dynamic balance. Simulation results show that this algorithm can improve use efficiency of system bandwidth resource notablly and increase admission ratio of high priority services patently.Applying complex network theory on multiple-constraint QoS multicast routing and its algorithm, constructing a minimum cost multicast tree than can satisfy multiple constraints, implementing different constraint performance requirements of different service to delay, jitter, packet loss ratio and bandwidth, reaching the goal of saving network bandwidth resource, decrasing network load and improving service quality. According to the characteristic of small-world networks and the preference of growth characteristic of BA scale-free network, the method for improving algorithm of PSO-CLIC based on comprehensive learning and intelligent cooperation is proposed. This algorithm divides a seed cluster into multiple child clusters, multiple child clusters carry evolution searching, thus increasing the algorithm's robustness and accuracy. Because of effect and constraint among the child clusters, the algorithm's capacity of searching optimum globally is improved. Based on generated network model, simulations to the multicast tree cost delay and jitter of the algorithm are carried, results show that this algorithm has a great advantage on convergence rate and solution accuracy.Applying complex network theory on research of malicious software spreading, a new malicious software spreading DP-SI(discrete probability susceptible infectious) model is presented, this modeling is decided by fixed characteristics of network and discrete probability of malicious software spreading. Selecting thress typical ER model of uniform network, NW model of non-uniform network and BA model of complex network model as simulation research objects, reserachs on malicious software spreading behaviors under the differnet topoloies are carried. Simulations show this model can be used to research malicious software spreading dynamics under any known toplogies, due to the modeling independent on some specific network structure. When some emergent events occur, this model can simulate spreading behavior swiftly, so it is beneficial to prevention and prediction. Besides, this model has flexible structure, so it is possible to change control strategies at any moment during evulotion process, while traditional differential equation model can do like this. At the same time, random immunization and destination immunization research on Internet is carried. Research results show that random immunization has no effect on Internet, while destination immunization is more effective, it can reduce malicious software spreading's speed greatly.Applying complex network theory on identifying specific P2P services, a P2P specific service flow identification method with tree-based multi-layer BP-LVQ neural network combination classifier is proposed. Comprehensive consideration on BP neural network and LVQ neural network, utilizing BP neural network as first layer classifier and LVQ neural network and second layer classifier, a tree model with two layer combination is formed, thus resolution efficiency, multi-services identification ability and anti-disturbance on data noise of classifier are improved. Specific P2P service identification is finished finally by means of LVQ neural network classification process. Through testing and analyzing on related network flows, unearthing flow characteristics and parameters, selecting optimum characteristic subset, forming P2P service classifier by applying BP neural network and LVQ neural network to make multi-layered combination, identification results can be gotten by associating specific P2P service confidence level. Testing results show that compared with P2P service identification method based on BP neural network, this identification method has great advantage on identification accuracy and identification speed.