Abstract:

This dissertation presents the design and implementation of evaluation frameworks that capture the applications or the user-perceived behavior in an integrated heterogeneous network model. An analysis of network systems within this framework is utilized for the design of cross layer interaction mechanisms wherein the protocols and applications can dynamically adapt to the changing environmental conditions.

A multi-paradigm WHYNET framework is introduced for in-situ evaluations of heterogeneous mobile systems. This framework is directed towards providing an accurate understanding of the protocol and application performance in the context of their interactions with system hardware and software, network architectures and wireless channel effects. To model heterogeneous networks with sensor components, the framework was extended with the SenQ simulator and emulator. SenQ offers the advantage of emulating a sensor operating system and protocol stack within a simulated environment. The analyses of network systems within these frameworks were applied in the development of cross-layer environmental adaptive network systems that adapt to wireless channel conditions and battery state. The dissertation has contributed to the wireless channel research by examining and proposing models for Environmental Mobility (EM) and in power management research by considering the impact of non-linear discharge and recovery effects of batteries.

Environmental Mobility refers to the ambient motion of entities like people, vehicles, etc. in the vicinity of wireless communication. Analysis with the channel measurement data concluded that the EM channel is a superimposition of multi-path fast fading and transient shadowing. The channel was modeled as a two-state Markov fading process and Fresnels theory of diffraction was extended to capture the transient shadowing. Analysis with the network protocols revealed that the EM channel has unfavorable implications for the classes of protocols that maintain state via feedback from the channel but are otherwise agnostic of the underlying operations. While these protocols are quick to learn the bad channel conditions, they are slow to discard those as conditions improve. A cross-layer scheme of Recovery from Earlier Good State (REGS) is presented that attempts to improve the performance of a wireless system when the channel conditions are amenable by discarding any memory acquired during the prior adverse channel condition durations. REGS was shown to be effective at multiple layers of the protocol stack.