MPI using Intel Fortran compiler

[adrian_ortega]

Hi everyone,

I have a question about how MPI works. Currently I have a job running on MPI on the mode master_slave, I mean that every slave executes a separate task without communicate with the others slaves, once a slave-node finishes, it sends a warning to the master and then the master delivers another job to the free slave.
Now, the master is in a do-loop with a parameter say k and I wish to deliver jobs from k_init to k_end; but here is the catch. By some strange reason, at the middle of the do-loop on k, one slave finishes running the task, but never it never warns the master that it finishes, I mean:

Call ProgramaCentral(kRank) ! Some slave node executes succesfully the task

Call MPI_Send( masterBuf, 3, MPI_INTEGER, 0, 0, MPI_COMM_WORLD, ierrMPI ) ! Some slave never warns the master

It appears that it get stuck in the limbo between these two lines.

 

[swartzism]

I have no experience working with MPI, yet. I'm going to guess that since you're using the combination of Fortran and MPI, you're working on some parallel programing. If that is the case, this question might be better directed at a HPPC forum (unless someone on here has some HPPC experience!) or someone who works at a supercomputing center.

 

[adrian_ortega]

Hi,

what does it mean HPPC? Is there on Physics Forums?

 

[swartzism]

HPPC means High Performance Parallel Computing. I'm not sure if Physics Forums has a spot for this. Cray might have a forum on the topic though. I don't mean to push you away from these forums -- that's not my intention.

There might be a wealth of information on what you need to know right on here!

 

[LudusRex]

Hello,

MPI (Message Passing Interface) is a parallel programming model used for distributed memory systems, where multiple processors or nodes communicate with each other through messages. The Intel Fortran compiler is a popular compiler used for scientific and technical computing.

In your case, you are using the master_slave mode in MPI, where each slave executes a separate task without communicating with the other slaves. This mode is useful for tasks that can be parallelized and do not require communication between nodes. However, it seems that one of your slave nodes is not properly communicating with the master, causing it to get stuck in the middle of the do-loop.

There could be several reasons for this issue. One possibility is that the slave node might be encountering an error or exception that is not handled properly, causing it to get stuck. Another possibility is that the MPI_Send function is not being executed correctly, which could be due to incorrect parameters or an issue with the MPI library.

To troubleshoot this issue, you can try debugging your code to see where exactly the slave node is getting stuck. You can also check the MPI documentation for the correct usage of the MPI_Send function and make sure that all the parameters are correctly specified.

I hope this helps. Good luck with your project!

Best regards,