A distributed mechanism for topology discovery in ad hoc wireless networks using mobile agents

• Authors: Romit RoyChoudhury;S. Bandyopadhyay;Krishna Paul
• Venue: MobiHoc '00 Proceedings of the 1st ACM international symposium on Mobile ad hoc networking & computing
• Year: 2000

The dynamics of wireless ad hoc networks as a consequence ofmobility and disconnection of mobile hosts pose a number ofproblems in designing proper routing schemes for effectivecommunication between any source and destination [1]. Theconventional proactive routing protocols that require to know thetopology of the entire network is not suitable in such a highlydynamic environment, since the topology update information needs tobe propagated frequently throughout the network. On the other hand,a demand-based, reactive route discovery procedure generates largevolume of control traffic and the actual data transmission isdelayed until the route is determined. Because of this long delayand excessive control traffic, pure reactive routing protocols maynot be applicable to real-time communication. At the same time,pure proactive schemes are likewise not appropriate for the ad hocnetwork environment, as they continuously use a large portion ofthe network capacity to keep the routing information current.In this paper, we have discussed a mobile multi-agent-basedframework to address the aspect of topology discovery in ad hocwireless network environment. In other words, we have designed amulti-agent based protocol to make the nodes in the networktopology aware. Our primary aim is to collect alltopology-related information from each node in the network anddistribute them periodically (as updates) to other nodes throughmobile agents. The notion of stigmergic communication [2]has been used through the implementation of a shared informationcache in each node. Moreover, we have used a concept of linkstability [3] and information aging based on which apredictive algorithm running on each node can predict the currentnetwork topology based on the current network information stored atthat node. We have demonstrated through performance evaluation of asimulated system that the use of mobile multi-agent framework wouldbe able to make each node in the network topology awarewithout consuming large portion of network capacity. This wouldeventually help us to implement a proactive routing protocolwithout much overhead. Moreover, as a direct outcome ofinfiltrating topology information into the nodes, the foundationsfor designing distributed network management and implementingcommunication awareness [3] get automatically laid.The basic idea of using agents for topology discovery has beenexplored in MIT Media Lab [4] earlier with certain limitations :first, node mobility and its effect on system performance has notbeen quantified. Second, the information convergence (convergenceof the difference between actual topology information and thetopology information as perceived by a node at any point of time)and its relationship with number of agents and agent migrationfrequency has not been clearly defined. Third, the navigationstrategies used do not ensure a balanced distribution of recenttopology information among all the nodes. We have tried to overcomethese difficulties. Moreover, we have defined a concept of linkstability and information aging based on which a predictivealgorithm running on each node can predict the current networktopology based on the current network information stored at thatnode.Mobile agents or messengers that hop around in the network [5]are a novel solution to the problem of topology discovery. Theagents hop from node to node, collect information from these nodes,meet other agents in their journey, interact with both to collectupdates of parts of the network that they have not visited or havevisited a long time back, and gift these collected data sets tonewly visited nodes and agents. It is to be noted that bycontrolling migration time-interval (time-to-migrate or TtM) of anagent, it is possible to control the agent traffic in the network.Moreover, the agent would always migrate from a node to only one ofits neighbor after pre-specified time-tick. So, the network wouldnever get flooded with propagation of agents.A major aspect underlying the infiltration of topologyinformation into mobile nodes is that the information carried mustbe recognized with a degree of correctness. Since the agentnavigation is asynchronous and there is an obvious time gap betweenthe procurement of information by an agent from one node and itsdelivery by the same agent to another node, it becomes imperativeto introduce a concept of recency of information. For example, letus assume two agents A1 and A2 arrive at noden, both of them carrying information about node m which ismulti-hop away from node n. In order to update the topologyinformation at node n about node m, there has to be a mechanism tofind out who carries the most recent information about node m :agent A1 or agent A2 ? To implement that,every node in the network has a counter that is initialized to 0.When an agent leaves a node after completing all its tasks at thenode, it increments that counter by one. We term this counter asrecency token. At any point of time, the magnitude of therecency token of any node represents the number of times that nodewas visited by agents since the commencement of the network. Thisalso implies that if two agents have a set of data concerning thesame node, say node A, then the agent carrying the higher recencytoken value of node A has more current information about it.We have defined a concept of information aging onlink-affinity based on which a predictive algorithm running on eachnode can predict the current network topology based on the currentnetwork information stored at that node. Link-Affinity ,associated with a link between two nodes n and m, is a predictionabout the span of life of that link in a particular context [3]. Weapply this notion to predict the topology by each node and thus becautious before data transfer is initiated.We have developed two metrics: average connectivityconvergence and average link-affinity convergence toquantify the deviation of actual network topology with the networktopology perceived by individual nodes at any instant of time. Wehave experimented with different mobility, transmission range andTtM on a 30-node ad hoc network. We have done a set of experimentsto derive that the optimum agent population should be half thenumber of nodes in the system. From the results, it is clearthat the average connectivity convergence improves with decrease inmobility. With TtM=100 msec., the connectivity convergence goesbelow 80 % for a high mobility of 30 m/s. However, the time tomigrate (TtM) could be lowered to produce better results even athigh mobility. But this has the obvious effect of congestion in thesystem as the system sees more agent traffic in the medium per unittime. However, our predictive mechanism in the context of TtM=100msec. could yield satisfactory results with the convergence valuesover 98 % (Fig. 1). Thus we see that resorting to the predictivemechanisms can improve performance significantly.