Applicability of Generic Naming Services and Fault-Tolerant Metacomputing with FT-MPI

Abstract

There is a growing interest in deploying MPI over multiple, heterogenous and geographically distributed resources for performing very large scale computations. However, increasing the amount of geographical distribution and resources creates problems with interoperability and fault-tolerance. FT-MPI presents an interesting solution for adding fault-tolerance to MPI, but suffers from interoperability limitations and potential single points of failure when crossing multiple administrative domains. We propose to overcome these limitations by adding “pluggability” for one potential single point of failure – the name service used by FT-MPI – and combining FT-MPI with the H2O metacomputing framework.