GossiPeer

• Description
• Keywords
• Architecture
• Development Team
• Documentation

 

Description

GossiPeer is a development framework for gossip-based communication protocols. It provides program designers with a toolkit for developing applications on a distributed set of machines (or nodes). GossiPeer is especially suited for large-scale deployment in dynamic settings since communication among distant nodes is based on gossip-based mechanisms. Its gossip (aka. epidemic) built-in protocols are, by essence, periodic and involve message exchanges between a constant number of neighbors.

Built upon this gossip communication paradigm, GossiPeer provides each node at the application level with a random set of nodes taken among all system nodes. This randomness is ensured by the implementation of algorithms that appeared recently in the literature and additional algorithms are currently under development.

GossiPeer is developed in Java for a compliance purpose, and allows deployment on world-wide distributed testbeds, e.g. PlanetLab, as well as on NFS distributed testbeds, e.g. EmuLab. At a lower level, gossip-based communication includes TCP or UDP, depending on reliability and speed requirements of the overlying application.

GossiPeer has already been used at IRISA, INRIA Futurs, and Cornell University for the development and deployment of three main distributed applications: Counting, Churn Measurement, and Distributed Slicing. The repository of the GossiPeer project is handled by the Inria Gforge.

Here is 4 page introduction [pdf]  

Keywords

• Gossip-based algorithms
• Peer-to-peer
• Epidemic algorithms
• Peer sampling
• Deployment
• Experimentation

 

Architecture Overview

Functionnal blocks  

Development Team

• Vincent Gramoli - Research Centre INRIA Futurs, Saclay.
• Erwan Le Merrer - Research Centre INRIA Rennes - Bretagne Atlantique.
• Anne-Marie Kermarrec - IRISA/INRIA-Rennes