Research On Topology Partitioning Oriented For Security Incident In Parallel Network Simulation

Network simulation has become an important means to research Internet, but the Internet's scale, complexity, heterogeneity and other issues seriously restrict the research on the Internet's behavior. Heterogeneous, complex, large-scale network simulation can't be achieved with single-machine because it requires vast computational resource. Parallel simulation technology provides a feasible method for large-scale network simulation. The major job of parallel simulation is task partitioning that is simulation tasks will be decomposed, and then multiple simulation nodes will be used to complete parallel simulation.The goal of task partitioning is to reduce network simulation time, increase the scale of simulation and improve the efficiency of simulation. Currently the most widely used approach for task partitioning is based on topology partitioning. Firstly, this thesis studies the mechanism of parallel network simulation, analyses the impact of task partitioning on various factors which determines the performance of parallel simulation, such as routing strategy, security incident types. Secondly,the thesis research on graph partitioning tool widely used, such as METIS of which the goal is to realize overload balancing then meet minimization of remote communication as far as possible. However, the traditional tool does not consider kinds of factors which impact the real network simulation performance, the result of partitioning is unreasonable. According to the above analysis, for the purpose of improving the routing efficiency, the thesis proposes an optimal partitioning algorithm based on subnet elimination that subnets are moved to adjacent and most appropriate other parts in allowable range. Comparing with the traditional algorithm, subnet elimination algorithm decreases the number of remote links and border routers, reduces the scale of remote routing tables, increases the speed of lookup routing tables and efficiency of packet forwarding, therefore it enhances the performance of parallel simulation.To further improve the simulation efficiency, two attack models are abstracted from security incident in simulation system. For these models, the thesis proposes two topology partitioning methods oriented for types of security incident, which is topology partitioning method applied to attack with target list that is TPBTL and topology partitioning method applied to attack with random scanning that is TPBRS. They calculate the load distribution of topological simulation process to define the weight of routers and links. Experiment results show that compared to the traditional partitioning algorithm, topology partitioning oriented for types of security incident can achieve higher simulation performance. Moreover it is more suitable for complex applications in large scale network environment.