HU Ruijing

PhD student at Sorbonne University
Team : REGAL
https://sites.google.com/site/ruijinghu1/

Supervision : Pierre SENS

Co-supervision : SOPENA Julien

Epidemic dissemination algorithms in large-scale networks: comparison and adaption to topologies

Information dissemination (broadcast) is essential for numerous distributed applications. This must be efficient, which limits the message redundancy, and ensures high reliability as well as low latency. We consider here the distributed algorithms that benefitting from the properties of the underlying topologies. Nonetheless, these properties and the parameters in the algorithms are heterogeneous. Thus, we should find a method to fairly compare them. First of all, we study the probabilistic protocols for information dissemination (gossip) executed over three random graphs. The three graphs represent the typical topologies of large-scale topologies: Bernoulli graph, the random geometric graph, and scale-free graph. In order to fairly compare their performance, we propose a new generic parameter: effectual fanout. For a given topology and algorithm, the effectual fanout characterizes the mean dissemination power of infected sites. Furthermore, it simplifies the theoretical comparison of different algorithms over one topology. After having understood the impact of topologies and algorithms on the performance, we propose an efficient reliable algorithm for scale-free topologies.

Defence : 12/02/2013

Jury members :

Aline Viana, Directrice de recherche, INRIA/Saclay, France [Rapporteur]
Xavier Defago, Professeur, JAIST, Japan [Rapporteur]
Matthieu Roy, Directeur de recherche, LAAS, France
Bertil Folliot, Professeur, UPMC, France
Julien Sopena, Maître de Conférence, INRIA/UPMC, France
Luciana Arantes, Maître de Conférence, INRIA/UPMC, France
Pierre Sens, Professeur, INRIA/UPMC, France
Isabelle Demeure, Professeure, Télecom Paris-tech, France

Departure date : 09/30/2014

2012-2013 Publications