During an emergency and rescue (ER) operation, like an earthquake, we often have no or only partially available communication infrastructure. One viable solution to enable data communication is to use a mobile ad hoc network (MANET). However, MANETs in ER scenarios are often disruptive in nature due to physical obstacles, distance and sparse node density. Therefore, we need delay tolerant networking (DTN) with a store-carry-forward mechanism, since we expect network partitioning. It is realistic to assume that some nodes travel between the network partitions, hence act as carrier nodes. However, assuming a priori knowledge about which nodes can act as carriers is unrealistic and unpractical.
In this master thesis, we analyze the application scenario and related work, and argue that using a prediction-based approach for detecting and utilizing carriers fulfills our requirements. We design a mechanism that provides dynamic selection of message carriers (DSMC) for DTN solutions. This is accomplished by that nodes in the network calculate, maintain and exchange delivery probabilities with all other nodes. Through this exchange, nodes learn which node has the highest probability of delivering the packets to the destination, hence act as carrier nodes. The design is implemented in NS3, together with the Dts-Overlay system, which is an ongoing development in the DT-Stream project to tackle network disruptions through an overlay. We evaluate DSMC and compare its performance against a carrier selection strategy, called Static-Dts, which relies on a priori knowledge of carrier nodes. Through extensive simulations, we show that, DSMC detects and utilizes carrier nodes for packet delivery. The performance of DSMC is nearly as good as Static-Dts, in terms of packet delivery. DSMC induces only a limited increase in delay, due to the dynamic selection of carrier nodes. Since the a priori knowledge required by Static-Dts is unrealistic, we argue that this increased delay is an acceptable trade off for a realistic solution. Through an analytical model, we show that the overhead of DSMC is low and scales well when the amount of nodes in a sparse MANET increases. For dense MANETs, this overhead potentially represents a problem of scalability, however these are not the kind of networks we are addressing.