Time Dependent Performance Analysis of Wireless Networks

Many wireless networks are subject to frequent changes in a combination of network topology, traffic demand, and link capacity, such that nonstationary/transient conditions always exist in packet-level network behavior. Although there are extensive studies on the steady-state performance of wireless networks, little work exists on the systematic study of their packet-level time varying behavior. However, it is increasingly noted that wireless networks must not only perform well in steady state, but must also have acceptable performance under nonstationary/transient conditions. Furthermore, numerous applications in today's wireless networks are very critical to the real-time performance of delay, packet delivery ratio, etc, such as safety applications in vehicular networks and military applications in mobile ad hoc networks. Thus, there exists a need for techniques to analyze the time dependent performance of wireless networks.\ud \ud In this dissertation, we develop a performance modeling framework incorporating queuing and stochastic modeling techniques to efficiently evaluate packet-level time dependent performance of vehicular networks (single-hop) and mobile ad hoc networks (multi-hop). For vehicular networks, we consider the dynamic behavior of IEEE 802.11p MAC protocol due to node mobility and model the network hearability as a time varying adjacency matrix. For mobile ad hoc networks, we focus on the dynamic behavior of network layer performance due to rerouting and model the network connectivity as a time varying adjacency matrix. In both types of networks, node queues are modeled by the same fluid flow technique, which follows flow conservation principle to construct differential equations from a pointwise mapping of the steady-state queueing relationships. Numerical results confirm that fluid-flow based performance models are able to respond to the ongoing nonstationary/transient conditions of wireless networks promptly and accurately. Moreover, compared to the computation time of standard discrete event simulator, fluid-flow based model is shown to be a more scalable evaluation tool. In general, our proposed performance model can be used to explore network design alternatives or to get a quick estimate on the performance variation in response to some dynamic changes in network conditions