Structures and Distributed Systems

The distributed system model is opposed to the traditional ”client/server” model because of the absence of a central server. Removing such central server has benefits since the whole network does not depend on a single point of failure anymore, the system is more robust. But this removal has a cost: the server centralizes data and has a global (full) knowledge of the system. Contrarily, in a distributed system, no entity has such a global knowledge of the system and its state. Each entity exchanges data with a subset of the system entities. We then speak about local (or partial) view of the system.

In this thesis, we study the structure induced by these local visions along with the impact of this structure on distributed applications. Our study will use different distributed system models: the partially synchronous model, the peer-to-peer model and the sensor network model. Our approach breaks down in three parts:

  • we show the impact of communication structures on distributed applications an show how to capture this impact.
  • we show how assumptions on the communication structure allow solving two traditional problems of distributed systems.
  • we show how to transform these communication structures into logical and explicit structures, in our case we build a coordinate system.

Composition du jury

  • Chairman of the board: Olivier Ridoux
  • Reviewers: Carole Delporte Gallet, Université Paris Diderot-Paris 7 LIAFA  &  Sébastien Tixeuil, Université Pierre et Marie Curie – Paris 6
  • Rapporteurs: Olivier Beaumont, LaBRI, domaine Universitaire  &  Michel Raynal, INRIA Rennes – Bretagne Atlantique
  • Supervisor: Achour Mostefaoui, Université de Rennes 1

Pd.D. Defense Gilles Trédan – November 26th 2009, 2.30pm

INRIA Rennes – Bretagne Atlantique, Salle Metivier