Backbone Communications Network, The Computer's Parallel Simulation Platform Developed

With the rapid development of internet technologies and applications, network simulation has become the method of choice for many computer network analysis problems. While carrying out large scale network simulation one quickly runs into two difficulties: first, the simulation of any reasonable size network can take excessive amounts of CPU time to execute; second, the memory utilized by a network simulation will grow at least linearly with the size of the network. Distributing the execution of the simulation over multiple processors offers one solution to alleviating these problems. There are generally two approaches to develop a parallel simulator. The traditional approach is to develop a simulator"from scratch", while the other involves interconnecting existing network simulators and reusing their established functionalities.In this paper, we designed and started developing a parallel network simulator following the latter approach. We interconnect several NS processes using MPI library for parallel execution and inter-process communication. Our work includes:(1) When designing the parallel simulator, we divide a network model into several sub models, which will be distributed on each processes of the simulator. We also defined the mechanism for inter-process communication and cooperation. We designed a multi-layer structure for each process of the parallel simulator. Each layer contains a set of modules of the simulator and is relatively independent with other layers, so that the complexity of developing and maintaining can be controlled.(2) While implementing the parallel simulator, we utilized several skills and methods and solved some key problems. We implemented ProcessTable module and a set of corresponding modules to initialize parallel parameters and to control execution flow. We defined an inter-process synchronization mechanism, and implemented UniformScheduler class. We inserted a serialization module in the packet data transmission process, which allows various packet data with complicated structure to be transferred between processes.(3) To order to allow functionality extensibility, we introduced Algorithm Process as well as its management module into the parallel simulator and defined the programming interface for Algorithm Processes. Self-defined algorithms such as network flow analysis algorithm can be added into the parallel simulator.(4) We carried out several simulation experiments using Internet2 network model and flow data. The simulation results show that the primary functionalities of the parallel simulator have reached our expected goals.