[mpich-discuss] I: Re: Problem to run MPICH2 on Windows 7 with user interface
Florin Ciubotaru
florinciu at gmail.com
Tue Apr 19 10:10:21 CDT 2011
Hi Rosa,
I think you need to have smpd running in a interactive console session as
opposed to a standard windows service. As Jayesh told me in a older thread,
smpd cannot run as a interactive windows service yet.
My particular issue was fixed by running smpd in debug mode as a user
process. For this, try stopping the smpd service on all nodes, then start it
manually from command line using "smpd -d 0". Run your regular mpiexec
command on the master node after that.
Hope this helps,
Florin Ciubotaru
On Tue, Apr 19, 2011 at 4:27 PM, Rosa Brancaccio <
rossella_brancaccio at yahoo.it> wrote:
> Hi all,
> first of all, thank you for your kind responses.
> I realize that I have explained myself very badly.
> I'm sorry, now I'll try to do it better :)
> I can run the "hello world" program (I copy the code below)
> without problems, both as traditional executable and as parallel
> program with n processes.
>
> #include "mpi.h"
> #include <stdio.h>
> #include <cstdio>
> #include <cstring>
>
> int main (int argc, char *argv[])
> {
>
> char txt={'\0'};
> int node;
>
> MPI_Init(&argc,&argv);
> MPI_Comm_rank(MPI_COMM_WORLD, &node);
> printf("Hello World from Node %d\n",node);
> fflush(stdout);
> printf("press any key to exit ");
> fflush(stdout);
> scanf(" ",&txt);
> fflush(stdout);
>
> MPI_Finalize();
>
> return 0;
> }
>
> I compiled this code both with Labwindows/CVI and Visual Studio 2010
> compiler, and I started the program by the command prompt (MSDOS windows).
> Of course, running the program as traditional executable the output
> is "Hello World from Node 0".
> Running the program as parallel executable (mpiexec -n 4 myapp.exe)
> the output is:
> "Hello World from Node 3"
> "Hello World from Node 0"
> "Hello World from Node 1"
> "Hello World from Node 2"
> in different order any time I run it.
>
> I tried to write the output on several files without problems.
> The program works fine both when compiled by Labwindows/CVI and
> by Visual Studio 2010. The program works fine both starting it as
> traditional executable and as parallel program (mpiexec -n 4 myapp.exe).
> The code is:
>
> int main (int argc, char *argv[])
> {
> FILE* f=NULL;
> char path[9999]={'\0'},name[99]={'\0'};
> int node;
>
> sprintf(path,"c:\\test\\file");
>
> MPI_Init(&argc,&argv);
> MPI_Comm_rank(MPI_COMM_WORLD, &node);
> sprintf(name,"_node%d.txt",node);
> strcat(path,name);
> f=fopen(path,"a");
> fprintf(f,"Hello World from Node %d\n",node);
> fclose(f);
>
> fflush(NULL);
> MPI_Finalize();
>
> return 0;
> }
> The output with n processes is n files with the correct written
> "Hello World from Node X" with X from 0 to n.
>
> The problem arises when I try to insert in my code the following line:
>
> MessagePopup("MPICH2 TEST","Test Successfully Completed \nOK");
>
> This line should start a small window with title "MPICH2 TEST" and
> the written "Test Successfully Completed OK" and a botton "OK" and
> the window should wait until I press the "OK" button and then it
> should expire.
> The "MessagePopup" command comes from LabWindows/CVI GUI Library and
> it works fine both when compiled by CVI and by Visual Studio.
> When I start this code as traditional executable the program write
> "Hello World from Node 0" and the window appears correctly.
> When I run the code with mpi (calling mpiexec -n 4 myapp.exe)
> unfortunately no writtens appear, no windows starts, and the
> command prompt is locked undefinitely until I press CTRL-C to
> kill the jobs. Moreover during this test the 4 myapp exist and I
> can see them in the list of windows task manager.
> I tried to use the -localroot options but nothing has changed:
> the program is locked and I must press CTRL-C to kill the processes.
> Anyway this is the messages (after the CTRL_C command):
> >mpiexec -n 2 -localroot MPI_test_VS.exe
>
> mpiexec aborting job...
> The thread to be suspended is no longer active, exit_code = 123
> unable terminate process safely. exit_code = 123
> unable to read the cmd header on the pmi context, Error = -1.
> Error posting readv, Connessione in corso interrotta forzatamente dall'host
> remoto.(10054)
> unable to read the cmd header on the pmi context, Error = -1.
> Error posting readv, Connessione in corso interrotta forzatamente dall'host
> remoto.(10054)
> job aborted:
> rank: node: exit code[: error message]
> 0: PC-ROSSELLA: 123: mpiexec aborting job
> 1: PC-ROSSELLA: 255
>
> At the end my question is:
> is there a way to use the CVI GUI library and MPI?
>
> Thank you all and best regards
>
> Rosa
>
>
> --- Lun 18/4/11, Jayesh Krishna <jayesh at mcs.anl.gov> ha scritto:
>
> > Da: Jayesh Krishna <jayesh at mcs.anl.gov>
> > Oggetto: Re: [mpich-discuss] I: Re: Problem to run MPICH2 on Windows 7
> with user interface
> > A: mpich-discuss at mcs.anl.gov
> > Cc: "rossella brancaccio" <rossella_brancaccio at yahoo.it>
> > Data: Lunedì 18 Aprile 2011, 19:55
> > Hi,
> > Did you try the "-localroot" option (mpiexec -n 2
> > -localroot myguiapp.exe)?
> > Also try running your program with mpiexec (myguiapp.exe)
> > . This might reveal error messages like missing dlls etc.
> >
> > Regards,
> > Jayesh
> >
> > ----- Original Message -----
> > From: "Nicolas Rosner" <nrosner at gmail.com>
> > To: mpich-discuss at mcs.anl.gov
> > Cc: "rossella brancaccio" <rossella_brancaccio at yahoo.it>
> > Sent: Monday, April 18, 2011 11:41:36 AM
> > Subject: Re: [mpich-discuss] I: Re: Problem to run MPICH2
> > on Windows 7 with user interface
> >
> > Ciao Rosa,
> >
> > > Unfortuntely my problem is not with I/O on files, but
> > with
> > > standard output and with user interface.
> >
> > If you're running into problems when trying to run the
> > simplest case
> > you've described (a basic "hello world" program that only
> > uses
> > stdout), may I suggest it could be a good idea to focus
> > exclusively on
> > that before mixing in additional issues?
> >
> > Although it is possible for an MPI-enhanced program to use
> > a GUI
> > library, it may not be trivial to get that right initially.
> > And, while
> > creating and writing to files is usually easier than GUI
> > stuff, it's
> > still not as trivial as simply writing to standard output.
> >
> > The latter is indeed trivial, and should be handled
> > automagically for
> > you, including the multiplexing of each rank's stdout (i.e.
> > output
> > should look as if your main mpiexec process had carried out
> > each
> > printf locally).
> >
> > If your simplest example still fails to work properly when
> > run under
> > mpiexec, could you please retry with a copy of the standard
> > "hellow.c"
> > code (see below) to see whether any problems persist?
> >
> > Regards,
> > Nicolás
> >
> >
> > #include <stdio.h>
> > #include "mpi.h"
> >
> > int main( int argc, char *argv[] )
> > {
> > int rank, size;
> > MPI_Init( 0, 0 );
> > MPI_Comm_rank(MPI_COMM_WORLD, &rank);
> > MPI_Comm_size(MPI_COMM_WORLD, &size);
> > printf( "Hello world from process %d of
> > %d\n", rank, size );
> > MPI_Finalize();
> > return 0;
> > }
> >
> >
> > >> I write the "hello word" test program. The MPICH2
> > library is linked
> > >> and there is no problem to run the program
> > (without call mpiexec)
> > >> from my compiler, obviously it starts with only 1
> > process and it
> > >> prints "Hello word from node 0" on stdout.
> > >> If I try to start the program from command line
> > (MS-DOS shell) with
> > >> the following command
> > >> mpiexec -n 4 Test.exe
> > >> 4 processes of test.exe start (I can see them in
> > the Windows Task
> > >> Manager) but the output does not appear, the
> > command prompt is
> > >> indefinitely locked until I press CTRL-C to stop
> > processes.
> > _______________________________________________
> > mpich-discuss mailing list
> > mpich-discuss at mcs.anl.gov
> > https://lists.mcs.anl.gov/mailman/listinfo/mpich-discuss
> >
> _______________________________________________
> mpich-discuss mailing list
> mpich-discuss at mcs.anl.gov
> https://lists.mcs.anl.gov/mailman/listinfo/mpich-discuss
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/mpich-discuss/attachments/20110419/a690cef2/attachment-0001.htm>
More information about the mpich-discuss
mailing list