« Topics in Delay Tolerant Networks (DTNs): Reliable Transport, Estimation and Tracking ».

le 12 Novembre 2012 à 15h30 à l’Université
Pierre et Marie Curie – Paris 6, salle 25-26/101, Jussieu.

Le Jury :

– M. Konstantin Avrachenkov, Rapporteur, INRIA Sophia Antipolis, France
– M. Tamer Basar, Rapporteur, University of Illinois, USA
– M. Guy Pujolle, Examinateur, Université Pierre et Marie Curie, France
– Mme. Lucile Sassatelli, Examinateur, Université de Nice Sophia Antipolis, France
– M. Eitan Altman, Co-Directeur de These, INRIA Sophia Antipolis, France
– M. Tijani Chahed, Co-Directeur de thèse, Telecom SudParis, France

Abstract:

Mobile Ad hoc NETworks (MANETs) aim at making communication between mobile nodes feasible without any infrastructure support. If the spatial density of mobile nodes in a MANET is low, then an end-to-end path between a source and a destination almost never exists; two mobile nodes can communicate only when they come within the radio range of each other. During the last few years, motivation
for development of MANETs has increased with the entry of intelligent devices with short range wireless communication methods. Sparse MANETs fall into the class of Delay Tolerant Networks (DTNs) which are intermittently connected networks and where there is no contemporaneous end-to-end path at any given time. The major or core research in DTNs addresses routing aspects while relatively fewer works exist on reliable transport; and when they do, they are mainly oriented towards deep space communication.

In this thesis, we first, propose a new reliable transport scheme for DTNs based on the use of ACKnowledgments (ACKs) as well as random linear coding. We model the evolution of the network under our scheme using a fluid-limit approach. We obtain mean file transfer times based on certain optimal parameters obtained through differential evolution approach. We account for the buffer expiry
time-out, quantify its impact on the optimal values of our protocol parameters and also demonstrate the adaptability of our optimal procedure to variations in expiry time-out.

Secondly, contact opportunities in DTNs are quite infrequent. This observation motivates the need to take the maximum benefit of these rare contacts. We thus propose and study a novel enhanced ACK approach called Global Selective ACKnowledgment (G-SACK) to improve reliable transport for DTNs covering both unicast and multicast flows. We make use of random linear coding at relays so that packets can reach the destination faster. We obtain reliability based on the use of so-called G-SACKs that can potentially contain global information on receipt of packets at all destinations. We obtain significant improvement through G-SACKs and coding at relays.

Finally, we tackle the problem of estimating file-spread in DTNs with direct delivery and epidemic routing. We estimate and track the degree of spread of a message/file in the network. We provide analytical basis to our estimation framework alongwith insights validated with simulations. We observe that the deterministic fluid model can indeed be a good predictor with a large number of nodes. Moreover, we use Kalman filter and Minimum-Mean-Squared-Error (MMSE) to track the spreading process and find that Kalman filter provides more accurate results as compared to MMSE.