Execution of MPI applications on Grid deployments suffers from node and network failure that motivates the use of fault tolerant MPI implementations. So, to ensure good performance in Grid, scalable fault tolerance should be taken into account. Research in fault tolerant MPI has led to the development of several fault tolerant MPI environments. The most popular approach is with coordinated checkpointing. Traditional coordinated checkpoint suffers from high storage and communication overhead because all processes must participate in the checkpointing at a time; also they all must roll back to the latest consistent global checkpoint even those processes that didn't fail. So in this paper, we propose a mailbox-based non blocking minimum-process coordinated checkpoint protocol (MNMCCP) for hierarchical Grid in which processes on different processors communicate indirectly by sending messages over the network through mailbox-based technique at a shared node. In this protocol, only the processes that communicated since last committed checkpoint will participate in the checkpoint at a time resulting in a reduction in stable storage access and enhanced performance. Our proposed protocol also exploits the mailbox of each process as an events logger since it logs the messages sent to the process in strict FIFO order. This combination of our coordinated protocol with message logging offers several additional advantages including limited computation lost, simplified recovery procedure and bounded recovery time because the effects of a failure are confined only to the processes that fail. From the other hand, using the proposed mailboxes technique ensures the reliable delivery of messages and prevents messages sent to the migrating or faulty process form losing and retransmitting. They will be kept in its associated mailbox, and the process will get them from the mailbox after restarting on the new resource. All of these make our protocol most suitable for highly dynamic environment where processes frequently migrate from one node to another thereby ensures the job to be executed within its deadline making such environment trust worthy.
