Reliable parallel programming model for distributed computing environments

Abstract

With the advent of large-scale heterogeneous platforms such as clusters and grids, resource failures are more likely to occur and have an adverse effect on the applications. Consequently, there is an increasing need for developing techniques to achieve reliability during execution. This paper presents FT-Jace, a new reliable programming model for grid computing environments. FT-JACE achieves reliability in a transparent manner for the programmer. It is based on active replication scheme, capable of supporting r arbitrary fail-silent (a faulty node does not produce any output) and fail-stop (no node recovery) node failures. The strength of our programming environment is that the deployment of the application does not require complicated mechanisms for failure detection. More precisely, node failures are masked and there is no need for detecting and handling such failures. We provide experimental results conducted on Grid'5000 platform to demonstrate the usefulness of FT-Jace. © 2010 Springer-Verlag.