A Channel Memory based fault tolerance for MPI applications

Fault tolerant message passing environments protect parallel applications against node failures. Very large scale computing systems, ranging from large clusters to worldwide Global Computing systems, require a high level of fault tolerance in order to efficiently run parallel applications. The Channel Memory approach provides the infrastructure for scalable tolerance to simultaneous faults. Along with a specially designed checkpointing system and recovery protocol, this approach has resulted in the MPICH-V architecture. In this paper, we describe CMDE - a stand-alone distributed program system based on MPICH-V architecture and implementing an approach to tolerate faults of Channel Memories.