Network Parallel Simulation Methods Research Based On NS3

As the computational complexity required in many research areas increases rapidly ordinary serial programs will reach the performance bottleneck,resulting in excessive simulation run time,can not fulfill people's growing demand,It is becoming increasingly important to improve the operational efficiency by using a variety of high-performance multicore hardware devices and parallel algorithms in parallel,the parallel programs not only reduce run time,but also across geographic location constraints,through the use of integrated types of computer equipment to save costs and also to increase the fault tolerance of the program.Network parallel simulation is one of the important applications.Through the network simulation we can research the running status of these networks and the impact of the entire network about the changes,this has a lot of important practical significance.This thesis mainly researches how to improve the efficiency of network simulation program through parallelism.Network simulation researches network behavior through network modeling.Quickly get the network running state,and the network changes to the impact of the entire network results for the network structure optimization and new business adjustment has a very important practical significance.So the use of parallel to reduce the simulation run time is necessary to study large complex network means.At present,there is a network simulation system developed with NS3 as a platform.With the expansion of the network topology scale,the simulation system improves the simulation efficiency through parallel simulation.However,due to some limitations of the partial implementation method,the parallel simulation efficiency is improved Can not meet the increase in the amount of data.In this thesis,some parallel methods in the simulation system are improved from three aspects to improve the simulation efficiency:1.For the original abnormal event scheduling module,when nodes increase and topology becomes more and more complex,as the realization of achieving an abnormal event,the global routing state update becomes extremely time-consuming,which directly affect the efficiency of parallel simulation.So we optimize the simultaneous simulation of the various types of abnormal events,to improve the efficiency of parallel simulation.2.The SNMP data to calculate the original partition algorithm related load is not strictly synchronized with the current task,This will lead to division of the results and the actual situation has a certain deviation,affecting the overall simulation results.Therefore,we use Netflow data to improve the network partitioning algorithm according to the inverse process of traffic matrix estimation,and then use the Metis algorithm to replace the original common KL algorithm by generating some link weight and node weight.3.According to the four steps of parallel mode design language,a generate ghost node to divide the traffic method is designed and implemented.This method is based on the parallel module of the original integrated data network simulation system.By judging the large traffic nodes,the corresponding nodes and related links are generated,and the traffic tasks are allocated to the original node and divided into half of the original node,so as to optimize the parallel simulation efficiency.The proportion to determine the large traffic nodes and the number of parallel division is the of the two key variables for the optimal efficiency.Through the experiment,the method can achieve more than 50% of the optimization efficiency.In this thesis,through the above three aspects to improve the simulation system,and using a network real collection of traffic tasks,the results of the test show that the parallel simulation performance and efficiency have been improved significantly.