CAREER: Designing Large-scale Ad Hoc Networking Systems: Models, Analysis, and Protocols

设计能够支持不同业务类型的大规模自组织网络系统涉及以非常复杂的方式交互的大量因素(例如，协议、参数、网络大小、业务特性)，使得系统优化成为一个不平凡的问题。这个项目的主要目标是加深对基本性能、扩展特性和权衡的理解，并包括两个主要研究主题。第一个重点是构建分析和经验模型，准确地描述性能指标(例如端到端延迟)和重要因素之间的函数关系。正在开发分析端到端延迟和分组丢弃模型，该模型考虑了业务到达过程、无线信道、差错控制机制和移动性引起的路径故障之间的交互作用。经验模型基于两种建模方法：(1)响应面方法和回归分析；(2)Levenberg-MarQuardt多层感知器神经网络。在第二个研究方向中，我们使用上述模型来探索几个可伸缩性和性能属性，包括哪些系统配置(S)在特定的感兴趣区域上产生最优、稳健、可扩展或可满足的性能响应(S)。利用预测经验模型，正在为传输和服务质量敏感的路由策略设计基于自适应跨层反馈的机制。拟议项目的成功将产生可靠的模型、原则和协议，在此基础上建立大规模的自组织网络。教育部分将通过将统计实验设计和建模研究纳入无线网络课程来加强教育，并将通过与非授予博士学位的大学建立研究伙伴关系来加强国家研究倡议。

Designing large-scale ad hoc networking systems capable of supporting heterogeneous traffic types involves a large number of factors (e.g., protocols, parameters, network size, traffic characteristics) that interact in a very complex manner, making system optimization a non-trivial problem. The principle goal of this project is to develop a deeper understanding of the fundamental performance, scaling properties, and tradeoffs and includes two main research thrusts. The first is focused on constructing analytical and empirical models that accurately characterize the functional relationship between performance metrics (e.g., end-to-end delay) and significant factors. Analytical end-to-end delay and packet discard models are being developed that account for interactions among the traffic arrival process, wireless channels, error control mechanisms, and mobility-induced path failures. The empirical models are based on two modeling approaches: (1) response surface methodology and regression analysis and (2) the class of Levenberg-Marquardt multilayer perceptron neural networks. In the second research thrust, we are using the above models to explore several scaling and performance properties, including which system configuration(s) lead to optimal, robust, scalable, or satisfiable performance response(s) over specified regions of interests. Using the predictive empirical models, adaptive cross-layer feedback-based mechanisms are being designed for transport and QoS-sensitive routing strategies. The success of the proposed project will produce solid models, principles, and protocols on which to build large-scale ad hoc networks. An educational component will enhance education by integrating statistical experimental design and modeling research into wireless networking courses and will enhance national research initiatives by establishing research partnerships with non-Ph.D. granting universities.