A multi-objective optimization of data dissemination in delay tolerant networks

• Published in: Expert systems with applications Vol. 57; pp. 178 - 191
• Main Authors: G Reina, D., Ciobanu, R.I., Toral, S.L., Dobre, C.
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 15.09.2016
• Subjects: Algorithms; Computer simulation; Data dissemination; Decision tree; Decisions; Delay; Delay tolerant networks; Expert systems; Genetic algorithm; Mathematical models; Multi-objective optimization; Networks; Optimization; Multi-objective optimization; Delay tolerant networks; Genetic algorithm; Data dissemination; Decision tree
• ISSN: 0957-4174; 1873-6793
• DOI: 10.1016/j.eswa.2016.03.038

•Data dissemination in DTNs as a multi-objective ptimization problem•Optimization of the hit rate, delivery cost, and delivery latency metrics•Set of non-dominated solutions given by the Pareto front•Selection of solutions from the Pareto front using decision trees•Social strength is the parameter with the highest influence on the balanced results Data dissemination in delay tolerant networks is an important issue due to the complexity of a multi-hop network formed by a high number of nodes with limited resources, variable and unpredicted mobility conditions. Nodes have to act as expert systems and make suitable forwarding decisions based on local knowledge on the fly. Most of the proposed algorithms rely on adjusting a range of decision variables related to social and topological aspects of the network. Adjusting such parameters is still an open issue since many of them are interrelated. To solve this problem, we propose a multi-objective evolutionary simulation framework for optimizing in terms of delivery hit, delivery cost and latency, a probabilistic data dissemination algorithm based on well-known and widely used social and topological parameters such as centrality, similarity, social strength, friendship, and trust. The proposed multi-objective based optimization framework provides many advantages with respect to existing approaches based on single objective optimization. Primarily, it allows the network designer to have a complete view of the possible outcomes of the data dissemination algorithm through the Pareto front (non-dominated solutions). Furthermore, we propose a decision tree-based selection to obtain under which values of the decision variables we can find a set of solutions that meet a target performance. We validate this selection mechanism by providing the conditions under which we can find balanced solutions in the considered simulation scenarios. The solutions provided by the proposed approach have significant implications for the design of new data dissemination algorithms in DTNs.