| Network evaluation tools are indispensable in design and developments of computer networks and their distributed applications. Traditionally, such tools have been developed based on an analytical model, discrete event simulation, emulation or a physical testbed of computer network. As computer networks grow in scale, heterogeneity and deploy a rich variety of new applications, none of the above four network modeling methodologies can fulfill the complex users requirements including scalability, fidelity, reconfigurability and experimentation support for real application, needed in the network evaluation studies.;In this dissertation, we developed a novel multi-paradigm network modeling framework, which address the limitations of existing evaluation tools by the integrations of modeling methodologies with complementary capabilities. As the first step, we developed MAYA, a modeling tool integrating analytical models, simulation, and physical network interface. MAYA allows users to interface simulated networks directly with physical networks, while attaining real time constraints even for large-scale networks by incorporating both analytical models and simulation. It thus provides a scalable and extensible evaluation tool for distributed applications to run on large, heterogeneous networks. To further enrich the modeling compatibilities of the proposed multi-paradigm framework, we developed TWINE, a hybrid emulation testbed that combines simulation, emulation and physical networks for evaluation of adaptive wireless systems, particularly those exploiting cross layer interactions. The integration of emulation and simulation in TWINE enables evaluation of real applications and protocols over a wide range of wireless network scenarios while achieving scalability of the testbed. Interfacing with real networks, TWINE also allows users to directly leverage existing, partially deployed physical testbeds and extend the performance evaluation to a larger-scaled networking context.;We have conducted experiments to evaluate the proposed multi-paradigm network modeling framework with respect to multiple performance metrics: Firstly, we evaluated the scalability improvements achieved by the integrations of multiple modeling methodologies; secondly, we evaluate the fidelity of the proposed framework in terms of accurately modeling the impact of various network dynamics on actual applications in real time; finally, we conducted case studies to demonstrate the utility of the proposed modeling framework in supporting repeatable, flexible and efficient experimentations with protocol implementations and applications in large heterogeneous networks. |
