Reducing Duplication of Messages in Gossip-based Dissemination Algorithms

Semester project

Gossip-based dissemination algorithms inherently duplicate messages to guarantee a very high level of reliability even in case a large number of peers fail. The duplication of messages has a cost in case of very reliable links, as many messages are received more than once by every peer.

We would like to experiment how one can reduce the duplication of messages by still guaranteeing a good reliability in the system. The solution we want to experiment with is to bundle, with the message, which peers have already received the message in order not to contaminate them again. In this very naive solution, the information for knowing “which peers received the message” will grow exponentially as big as the system, which we want, of course, to limit.

We want thus to experiment “what kind of information” to bundle, finding an optimal trade-off between the size of information to bundle (thus the overall bandwidth used) and the number of total duplicated messages in the system.

Language: Java

Responsible: Maxime Monod

Project intended for 1 student

Le superviseur parle aussi français…