Develop Of A Parallel Simulation Scheduler For NS2 Over A Cluster Of Computers

The networks simulation is an important and very helpful means for both network design and performance analysis. With the simulated network's scale increasing and the network-simulation performance bottlenecks appearing, the technology of parallel simulation is increasingly popular. This technology uses multiple processing units to execute simulation program in parallel, removing limitations of the serial emulation on the network scale and simulation time, becoming an active research topic of the network. This thesis focuses on the design and implementation of the parallel simulation speedup for wireless network.This thesis firstly describes the technology of parallel computing and the open source implementation of the Message Passing Interface standard, OpenMPI, including the architecture and the application programming interface. The architecture and serial event scheduler of NS2 network simulator is also discussed. Then, the thesis analyzes the methods of implementation for parallel simulation, such as task parallelization, splitting, and allocation, and virtual time synchronization.Then, this thesis analyzes the programming requirement of a parallel simulation of wireless networks, including task assignments, virtual time synchronization, and cross events synchronization. After that, the design and implementation proposal of the parallel simulation is given, including the design of local schedulers and the approach of tasks assignment and mapping among parallel processors. The realization of virtual time synchronization, events synchronization and trace files out-of-order issue of event-tracing among distributed computing is also discussed in detail.Finally, this thesis designs two types of test cases, small-scale simulation scenarios and large-scale simulation scenarios, respectively for the verification of simulation conformance and the test of speedup efficiency. The test cases are executed by multi-node-multi-processor, which is compared to the serial execution, in order to test simulation conformance and speedup efficiency of the parallel simulation scheduler. The measured speedup of three large-scale simulation scenarios are 10.7, 12.7 and 48.3, respectively.