الفهرس | Only 14 pages are availabe for public view |
Abstract Network Simulation is an essential step in evaluating and validating invented network protocols. However, all of the existing network simulators are not scalable to large-scale network topolo- gies in terms of hardware resources (RAM and CPU speed) and the taken simulation time. A reasonable solution for such a scalability problem is to partition the simulation experiment on multiple physical machines. This thesis develops and proposes a topology mapping solution to partitions large-scale topologies and map the created partition to separate physical nodes. The developed topology mapping is integrated into the Collaborative Radio Cloud (CRC) to improve the scalability of the testbed. The testbed is enabled to support both manual topology mapping experiments and automatic topology mapping experiments. A genetic algorithm based solution is proposed to implement the automatic topology map- ping (GA-based mapping). The proposed GA-based mapping aims to reach the minimum pos- sible simulation time for a large-scale experiment, in addition to utilizing the used physical machines. Furthermore, it fulfills the primary consideration of a mapping solution, of keeping the evaluation experiment outside the influence of the applied partitioning. The experimental evaluation results of the proposed GA-based mapping proves its out- weighed performance over two mapping techniques: Graph cut mapping and random search mapping. Moreover, the GA-based mapping achieves a reduction in simulation time reaches 90% while keeping the utilization of all used machines above 95%. Hence, the proposed map- ping technique fulfills the desired goals. |