I would suspect you are hitting limits on the number of open sockets - check 
your limits settings on file descriptors.

On Jun 1, 2013, at 11:43 AM, vacate <vacatehop...@gmail.com> wrote:

> Hello everybody,
> 
> I'm a beginner in Open MPI world.
> Maybe it's a simple problem, but I cannot figure out what happen on it...
> 
> My situation is
> I use 4 hosts totally, and their IP address are static.
> I can't do mpirun over 1500 times almost at the same time.
> (but it's always okay less than 1000 times)
> I got many "ssh_exchange_identification: connection closed by remote host" 
> errors.
> 
> --------------------------------------------------------------------------------------------------------------------------
> My Open MPI version : 1.6.2
> --------------------------------------------------------------------------------------------------------------------------
> I use a simple bash shell script file to run my Open MPI file(named 
> openMPI_test)
> Here is my script content :
> 
> for (( index=0; index<2000 ; index++))
>    do
>        (time mpirun --hostfile my_hostfile openMPI_test &) >> file 2>&1
>    done
> 
> 
> p.s.1 "my_hostfile" file lists 4 hosts' IP address.
> p.s.2 "openMPI_test" file ask each host do the same thing, it means:
>           if(rank == 0){      for(i=0 ; i<65535 ; i++)    temp = i/(i+1);  }
>           else if(rank == 1){      for(i=0 ; i<65535 ; i++)    temp = 
> i/(i+1);  }
>           else if(rank == 2){      for(i=0 ; i<65535 ; i++)    temp = 
> i/(i+1);  }
>           else if(rank == 3){      for(i=0 ; i<65535 ; i++)    temp = 
> i/(i+1);  }
> --------------------------------------------------------------------------------------------------------------------------
> 
> At the first, I thought I have some system problems,
> so I tried to modify my /etc/ssh/sshd_config file.
> I set Max_Sessions up to 65535, and MaxStartups up to 65535,
> but the result made me so sad because it still didn't work :((
> 
> I'm not sure if there are some settings or limits in Open MPI,
> or I just have another system problems?
> 
> I really hope someone can help me..
> Thank you all very very much!!
> 
> 
> 
> Best Wishes,
> Jen
> 
> _______________________________________________
> users mailing list
> us...@open-mpi.org
> http://www.open-mpi.org/mailman/listinfo.cgi/users

Reply via email to