Analysis of an agent-based metric-driven method for ad-hoc, on-demand routing

Ad-hoc networks could bring many benefits over traditional, fixed networks. Unfortunately, they have many difficulties, including the selection of the optimal route, the availability of routes as mobile nodes move around the physical space, and problems related to the routing capabilities of mobile devices. This paper outlines an agent-based method which allows tests to be run on nodes within an ad-hoc network, which determines their fitness to route data. These tests are related to important routing parameters, such as data processing performance, memory buffering capabilities, system utilisation, network performance, and so on. A key element is the usage of the battery on mobile networks, and the performance of an ad-hoc network can be affected by mobile nodes leaving the network due to lack of battery power. The paper outlines the tests that could be used on devices, and presents results which can be used to assess the fitness of a device to route data through an ad-hoc network. The overall architecture uses a three-layered model where static agents gather the fitness information, which is then passed to domain agents, which are either static or mobile agents, who are then responsible for collecting the information for inter-domain routing agents, who are responsible for electing gateway agents. A key factor in the determination of routing metrics is the performance of the device, where a device could be used to route certain types of data, based on their routing performance. The results show practical results on how well a range of devices respond to important tests, such as memory buffering capabilities, network performance and battery usage. This is especially important when limited-power devices, such as PDAs, are used in the ad-hoc network. The results also show, clearly, how a typical iPAQ PDA copes with these tests, against high-powered PCs. An important result is in battery usage which shows that it varies with the type of processing and network activities that is being performed on the device. From these results reliable models could be built which predicts battery usage for differing types of activities, especially in routing data. 2004 Elsevier B.V. All rights reserved.