Distributed Asynchronous Iterative Algorithms: New Experimentations with the Jace Environment

Jacques M. Bahi, Raphaël Couturier, David Laiymani, Kamel Mazouzi

Abstract: Jace is a Multi-threaded Java environment that permits to implement and execute distributed asynchronous iterative algorithms. This class of algorithm is very suitable in a grid computing context because it suppresses all synchronizations between computation nodes, tolerates the loss of messages and enables the overlapping of communications by computation. The aim of this paper is to present new results obtained with the new improved version of Jace. This version is a complete rewriting of the environment. Several functionalities have been added to achieve better performances. In particular, the communication and the task management layers have been completely redesigned. Our evaluation is based on solving scientific applications using the french Grid’5000 platform and shows that the new version of Jace performs better than the old one.