Spanning Trees from theory to practice. Self-Stabilizing algorithms and sensor networks

Les réseaux de capteurs sont des réseaux particuliers composés d'objets contraints en ressources. Ils possèdent une faible puissance de calcul, une faible puissance de transmission, une faible bande passante, une mémoire de stockage limitée ainsi qu'une batterie à durée de vie limitée. Afin d'intégrer de tels réseaux dans l'internet des objects, de nouveaux protocoles ont été standardisés. Parmi ces protocoles, le protocole RPL (pour Routing Protocol for Low Power and Lossy Networks). Ce protocole est destiné a construire une topologie logique de routage appelée DODAG. Dans cette thèse, nous abordons l'aspect acheminement de données qui considère une topologie de routage arborescente. L'acheminement des données se fait donc de saut en saut d'un enfant à son parent (ou d'un parent à son enfant). Optimiser la construction du DODAG revient donc à construire un arbre couvrant selon une contrainte donnée. Un arbre couvrant est une structure communicante qui permet de maintenir un unique chemin entre toutes paires de noeuds tout en minimisant le nombre de liens de communication utilisés. De plus, nous considérons les contraintes des réseaux de capteurs telles qu'une batterie déchargée et la variabilité du lien radio comme des fautes transitoires. Ceci nous conduit par conséquent à construire une structure couvrante tolérante aux fautes transitoires. L'auto-stabilisation est une branche de l'algorithmique distribuée qui assure qu'à la suite d'une ou de plusieurs fautes transitoires, le système va retrouver de lui-même un comportement correcte au bout d'un temps fini. L'objectif de cette thèse est de proposer des algorithmes auto-stabilisants dédiés aux réseaux de capteurs.Spanning Trees from theory to practiceSelf-Stabilizing algorithms and sensor networksAbstract : Sensor networks are composed of ressources constrained equipments. They have low computing power, low transmission power, low bandwidth, limited storage memory and limited battery life.In order to integrate such networks in the Internet of things, new protocols were standardized such as RPL protocol (for Routing Protocol for Low Power and Lossy Networks). This protocol is intended to build a logical routing topology called DODAG (for Destination Oriented Directed Acyclic Graph). In this thesis, we discuss the data routing aspect by considering a tree routing topology. Thus, the routing of data is hop by hop from a child to its parent (or from a parent to its child). Optimize the construction of the DODAG is therefore to build a spanning tree in a given constraint. A spanning tree is a connecting structure that maintains a unique path between all pairs of nodes while minimizing the number of used communication links. Furthermore, we consider the constraints of sensor networks, such as a dead battery and the variability of the radio link as transient faults. This leads us to build a covering structure tolerant to transient faults. The self-stabilization is a branch of distributed algorithms that ensures that following one or more transient faults, the system will find itself a correct behavior after a finite time.The objective of this thesis is to propose self-stabilizing algorithms dedicated to sensor networks. The contributions of this thesis are:In the first part of the thesis, we proposed a self-stabilizing algorithm for the construction of a minimum diameter spanning tree.This construction is natural when we want to minimize the communication delay between a root and all other network nodes. Our algorithm has several advantages. First, our algorithm is limited to memory occupation of O(log n) bits per node, reducing the previous result of an n factor while maintaining a polynomial convergence time. Then, our algorithm is the first algorithm for minimum diameter spanning tree that works as an unfair distribution demon. In other words, we make no restriction on the asynchronous network behavior. In the second part of the thesis, we are interested in the unstable topology built by RPL protocol (DODAG). Our solution is to place an additional constraint on the number of children a node can accept during the construction of the DODAG. This constraint has the effect of reducing the rate of parent change and consequently to improve the protocol performance in terms of packet delivery rate, delay of communication and power consumption. In addition, we implemented a mechanism to update the information of the downward routes in RPL. Furthermore, our solution has the advantage of not generating overhead because we use existing control messages provided by RPL to implement it. Finally, this contribution is twofold since we validated our solution both by simulations and experiments