Component Based Modeling for Cross-layer Analysis and Design of MAC and Routing Protocols in Ad-hoc Networks

J.S. Baras, V. Tabatabaee, and K.S. Jain

To appear in the Proceedings of MILCOM 2009 "The Challenge of Convergence", Boston, Massachusetts, October 18-21, 2009.

Abstract

A large number of routing protocols have been proposed for MANET. However, todate most of performance studies are based on discrete event simulations, with very few exceptions. Furthermore, all routing protocols are essentially large monolithic software, which are very difficult to adapt to varying conditions and mission scenario. These simulations are very time consuming and provide little insight on design parameter sensitivities or on how we can improve performance and adaptivity of these routing protocols. We present in this paper, a complete scenario driven component analytic model of MANET routing protocols and we use it to systematically study their performance. Under the proposed component based model the routing protocol is broken up into different components, whose performance although coupled, can be independently evaluated and designed for better performance over a wider scenario settings. For example, OLSR is broken-up into Neighborhood Discovery Component (NDC), Selection of Topology for Dissemination Component (STIDC), Topology Dissemination Component and Route Selection Component (RSC). Further, the performance of these components depends on the underlying scheduler, MAC and PHY layers. ’ As a specific example we present a performance study of the OLSR protocol in MANET networks using 802.11 MAC. In several scenario driven studies, with user-specified topologies and traffic demands, we study the performance metrics - throughput, delay and packet loss, for each connection and of the overall routing protocol. We generalize the NDC component of OLSR, for better control over the state of links or neighbors. Using Automatic Differentiation techniques, we quantify the performance tradeoffs between certain parametric choices in each of the components. Lastly, we validate our performance models using network simulations.