Authors

Samir Jafar, Sebastien Varrette, Jean-Louis Roch

 Abstract

In this paper, to achieve correct execution of peer-to-peer applications on non-reliable resources, we present a portable and distributed algorithm that provides fault tolerance and result checking. Two kinds of faults are considered: node failure or disconnection and result forgery. This algorithm is based on the knowledge of the macro dataflow dependencies beween the application tasks. It provides correct execution with respect to a probabilistic certificate. We have implemented it on top of Athapascan programming interface and experimental results are presented.

Keywords

Fault Tolerant, Checkpoint Recovery, Certificate of execution, Result Checking, Parallel Processing