Plagemann et al. Cross-layer Overlay Synchronization in Sparse MANETs Cross-layer Overlay Synchronization in Sparse MANETs Thomas Plagemann, Katrine S. Skjelsvik, Matija Puzar, Aslak Johannessen, Ovidiu Drugan, Vera Goebel, Ellen Munthe-Kaas Department of Informatics, University of Oslo {plageman, katrins, matija, aslakjo, ovidiu, goebel, ellenmk}@ifi.uio.no ABSTRACT Mobile Ad-Hoc Networks maintain information in the routing table about reachable nodes. In emergency and rescue operations, human groups play an important role. This is visible at the network level as independent network partitions which are for some time stable before their members change through merging or partitioning. We use the information from stable routing tables to optimize the synchronization of Mediators in a Distributed Event Notification System. In a stable partition each node has the same information, thus a single Mediator can efficiently coordinate the synchronization, while all other Mediators just receive updates. We show in our experiments that just a few seconds are needed until routing tables stabilize and all nodes have a common view of the partition. We present a heuristic to determine the proper time to synchronize. Furthermore, we show how exceptions, like disappearing coordinating Mediators and unexpected messages, can be efficiently handled. Keywords Synchronization, sparse MANETs, cross-layer optimizations, overlays in MANETs, publish-subscribe. INTRODUCTION In order to efficiently handle crises and emergencies, emergency and rescue (ER) teams benefit from well working communication infrastructures for command, control and coordination. However, first responders are typically confronted with an environment where no communication infrastructure is available, either because it was nonexisting, or the earlier existing ones have been destroyed. Therefore, wireless Mobile Ad-Hoc Networks (MANETs) formed by devices carried by ER personnel are often the only means to establish a communication infrastructure. However, the mobility of the ER personnel combined with the size of the emergency area (which is typically multiple times larger than the coverage of individual IEEE 802.11 radios) and obstacles in the area reflecting radio waves, leads to the situation that there is often not one single MANET connecting all ER personnel. Instead, multiple partitions might exist and change over time through merging and partitioning. Typically, these partitions correspond to groups of ER personnel that have a common task to fulfill. Due to the dynamics of ER operations, groups might need to change their locations, and group memberships might change. This is reflected at the network level through changes in the routing table. Evidence for such group mobility is not only given by our study of ER operations, but also confirmed by recent studies of social mobility [2] and community detection [6] in opportunistic networks. We have developed a Distributed Event Notification Service (DENS) [10] to support remote patient monitoring. DENS uses Mediators to replicate subscriptions, to enable graceful degradation in case of network partitions, and to convey subscriptions and notifications from source to destination. If there is connectivity to the destination the Mediator uses the OLSR MANET routing protocol [4] and IP to transport the packets to their destination. However, if a destination node is turned off or in a different partition, OLSR and any other MANET routing protocol fails. Therefore, the Mediators form an overlay on top of the MANET to perform delay tolerant transport through so- called “store-carry-forward” operations [10]. The replication of undelivered subscriptions and notifications increases the probability that one of the Mediators at a later point in time can join a partition with formerly unreachable destinations. The number of Mediators can be dynamically adapted to the particular ER operation and by this trade availability of DENS against resource consumption, i.e., the more Mediators the higher the availability and the higher the resource consumption. The message ferry approach [12] is determining the mobility, including speed and trajectory of special nodes called ferries, to make sure that a previously unreachable destination and the ferry are coming into communication range. This is not in general possible in ER operations. Therefore, we replicate undelivered subscriptions and notifications Proceedings of the 5 th International ISCRAM Conference – Washington, DC, USA, May 2008 F. Fiedrich and B. Van de Walle, eds. 546