Re: questions about checkpoint/restart on multiple clusters of MPI

From: Paul H. Hargrove (PHHargrove_at_lbl_dot_gov)
Date: Mon Mar 22 2010 - 15:19:18 PDT

  • Next message: fengguang tian: "Re: questions about checkpoint/restart on multiple clusters of MPI"
    fengguang,
    
    As an author of the BLCR software, I cannot guess from your question 
    what the problem might be.
    
    You say "the MPI program are not terminated as what happened in single 
    machine", but you don't indicate if you used the "--term" argument in 
    either the single machine or cluster case.  One should not expect the 
    program to terminate immediately in either case unless "--term" is 
    passed (to cr_checkpoint or ompi-checkpoint, respectively).
    
    You say "ompi-restart also doesn't work on cluster", but you don't 
    provide any information on how it is failing.  Is there any error 
    message from ompi-restart?  Does the ompi-restart command exit, or does 
    it hang?  Unless OMPI is doing something special, I think that if the 
    original MPI application is still running when you try to restart, then 
    a restart failure is the expected behavior because BLCR is going to try 
    to allocate the original PIDs.
    
    Since your problems sound specific to your use of MPI, I suggest that 
    your best chance of getting a good answer would be to ask on the 
    ompi-users mailing list.  You can find out about the Open MPI mailing 
    lists at www.open-mpi.org.
    
    -Paul
    
    fengguang tian wrote:
    > I set up a cluster of 18 nodes using Open MPI and BLCR library, and 
    > the MPI program runs well on the clusters,
    > but how to checkpoint the MPI program on this clusters?
    >
    > what I have done is that: I run the program using mpirun in the shared 
    > directory on the master node, and use
    > ompi-checkpoint command in another terminal on master node, it will 
    > create an checkpoint file,but the MPI program
    > are not terminated as what happened in single machine. Also, the 
    > ompi-restart also doesn't work on cluster.
    >
    > what should I do ?
    >
    > Cheers!
    > fengguang
    
    
    -- 
    Paul H. Hargrove                          PHHargrove_at_lbl_dot_gov
    Future Technologies Group                 Tel: +1-510-495-2352
    HPC Research Department                   Fax: +1-510-486-6900
    Lawrence Berkeley National Laboratory     
    

  • Next message: fengguang tian: "Re: questions about checkpoint/restart on multiple clusters of MPI"