Avoiding network-wide broadcasting with controlled flooding for on-demand ad hoc routing protocols Lu´ ıs H. M. K. Costa, Marcelo D. de Amorim, and Serge Fdida LIP6 – Universit´ e de Paris VI 8, rue du Capitaine Scott 75015 – Paris – France Email: costa,amorim,sf @rp.lip6.fr. Abstract—Ad hoc routing protocols that use broadcast for route discovery may be inefficient if the path between any pair source- destination is frequently broken. In order to efficiently use net- work resources, this paper proposes to use a controlled flooding (CF) mechanism to replace network-wide broadcast in route dis- covery procedures. CF is used to build alternative routes around the shortest-path between the source and the destination. With al- ternative routes, data packets are forwarded to a secondary path without requiring the source to re-broadcast the network. Fur- thermore, these alternative routes serve as guidelines indicating an approximative direction for future controlled floodings. CF is not orthogonal to the route discovery algorithm, but only complemen- tary. Indeed, CF requires a pre-existing path to operate. The key idea is that the source only performs a new broadcast if all paths (shortest path and alternative paths) are lost. CF can be used for many purposes. We detail in this paper one of the applications of controlled flooding: connectivity maintenance. We show that our proposal reduces the connection disruption probability as well as the need for network-wide broadcasts. Index Terms—Controlled flooding, alternative routing, connec- tivity maintenance, broadcast. I. I NTRODUCTION Typical on-demand ad hoc routing protocols use broadcast to discover routes [1], [2], [3]. In spite of providing fast route dis- covery, broadcast has several inconveniences. In mobile ad hoc networks, wasting resources is specially undesirable, because these resources are often very limited. Several mechanisms have been proposed to improve the performance of broadcast- based protocols, like interception of request messages, prede- cessor advising, and limited flooding [4], [5], [6], [7]. Nev- ertheless, broadcast is always performed whenever a node re- quires a path to an unknown destination. In on-demand routing protocols, for instance, a node typically maintains a route to a specific destination until it receives a path error message from downstream nodes or a specific timer goes off. If a message has no means to reach the destination, the source performs a new broadcast to discover a new route. Broadcast is difficult to avoid when the source has no idea of the current location of the destination. Nevertheless, broad- casting the entire network for each path discovery is inefficient. We argue in this paper that broadcast should be performed for the first search, but can be avoided for some of the subsequent ones. Suppose that the source, , has a route to the destination, . If after some time this path is broken, it is possible that is not (very) far from the original location if node mobility is lim- ited. Performing broadcast to discover another path is not the best solution because some nodes (at least the ones that were in the previous path) have a clue of the location of . In this case, the search can be “directed” toward the previous position of the destination, without flooding the entire network. We address in this paper the connectivity maintenance prob- lem in on-demand ad hoc routing protocols. When a path is broken, data must wait until another path is constructed. If mo- bility is high, the communication between a source-destination pair may be frequently stopped, which is prejudicial for delay- sensitive applications. The ideal configuration would be to promptly forward the messages through another path. In proac- tive protocols, this is straightforward. Nevertheless, proactive approaches are efficient for small populations, but do not scale for larger ad hoc networks. In the case of reactive (on-demand) protocols, additional mechanisms have to be provided since we have a priori only one path per active destination. This paper proposes the use of controlled flooding (CF) to in- crease the probability that a source find a path to a specific desti- nation without requiring the broadcast of request messages. Be- sides the shortest path ( ) between the source and the des- tination , the network also establishes some alternative paths “around” . The role of the alternative paths is twofold. First, data packets sent to the destination do not have to be queued until a new path is discovered. As soon as a node detects that its successor toward the destination has failed, it promptly forwards data packets to one of the alternative routes. Second, nodes in alternative paths keep reduced state that is useful for future path discovery. This state indicates in which direction to start a search procedure. This avoids the source to (re)flood the entire network, but instead do a controlled flooding, possi- bly saving network resources. In this paper, we consider only the first characteristic. We will show that in many situations the source does not have to restart the discovery procedure, which leads to a significant reduction of the number of control mes- sages in the network.