Inter-relationships of performance metrics and system parameters in mobile ad hoc networks

In this paper, the network behavior of a routing algorithm for mobile ad hoc networks is investigated. Extensive simulations are performed using ns-2 in a variety of mobility scenarios and offered load regimes. In the literature, performance metrics (goodput, delay and path length) are often obtained through ensemble averaging of many flows. Here we advocate an alternate graphical interpretation of simulation results similar to that used by Holland and Vaidya (see Proceedings of the fifth annual ACM/IEEE international conference on Mobile computing and networking, August 1999). Performance metrics of individual monitored flows are plotted instead. By identifying the correlations between performance metrics and system parameters, inter-relationships between them are revealed. For example, we have shown that path length is dependent on system parameters such as mobility, offered load and even the node distribution. These observations often give us insights to the mechanisms that underline the network behavior. In particular we have resolved a conjecture that goodput improvement under high mobility is due to the load balancing effect. We show that at high mobility, goodput improvement for heavy offered load regimes is a consequence of the reduction of path length in the flows. Furthermore, we have introduced the concept of fraction of congested flows as a new performance metric. This and some other metrics such as fairness can be visualized from our graphs and are important in characterizing network performance.