Hi, I have 3 questions to ask about,

 

1, how does open-mpi find the faulty node?

 

2, if one node is dead, could the programs continue running? How about two nodes or even more nodes are dead ?

 

3, How to recovery faulty node (dead node) ? Is there any possibilities to recover without check-pointing, since it is time-consuming and decrease performance ?

 

Thanks!

 

Rui Wang

 

ICT, P.R. China