Study On The Scalability And Fidelity Of Network Simulation And Emulation Platforms

As the scale of network continues to expand, researchers have an increasingly higher demandfor large-scale network experiment platform. Due to the enormous cost of building a real networktestbed, network simulation and emulation platforms have been widely used in network research.This thesis focuses on the scalability and fidelity of these two platforms to meet the above needs.Based on NS2network simulation platform, a cache-aware scheduling algorithm is designed toimprove its computing performance and scalability by simulation acceleration. The implementationof enqueue and dequeue operation is described in detail. A theoretical analysis shows that theenqueue and dequeue operation speedup of the proposed algorithm with the default calendaralgorithm is proportional to the ratio of caching to memory in rate. Cache hit rate is measured todirectly observe the cache utilization effectiveness through the tool pfmon. Experiments of anetwork with100nodes are conducted. It is shown that the cache-aware algorithm is morebeneficial in terms of cache utilization and can speed up the simulation by a factor of29inmaximum.In this thesis, we also present a design of network emulation platform which integrates opensource packages of GNS3, Xen and NetEm. The virtualization functions of Xen and GNS3are usedto develop the scalability of the platform. The fidelity is guaranteed by using real protocol stacks forterminals, employing NetEm for links, and applying GNS3for routers. Experiments with thedumbbell model accurately reflect the fairness, friendliness and congestion window growthcharacteristics of CUBIC and Reno algorithm so as to verify the feasibility and fidelity of platform.