Hi Xianjun

Are you running on two processes (mpiexec -n 2)?
I think this code will deadlock for more than two processes.
The MPI_Recv won't have a matching send for rank>1.

Also, this is C, not MPI,
but you may be wrapping into the negative numbers.
Have you tried to print Gsize?
It is probably -2147483648 in 32bit and 64bit machines.

My two cents.
Gus Correa

Mike Dubman wrote:

Hi,
What interconnect and command line do you use? For InfiniBand openib component there is a known issue with large transfers (2GB)

https://svn.open-mpi.org/trac/ompi/ticket/2623

try disabling memory pinning: http://www.open-mpi.org/faq/?category=openfabrics#large-message-leave-pinned


regards
M

2010/12/6 <xjun.meng@gmail.com <mailto:xjun.meng@gmail.com>>

   hi,

   In my computers(X86-64), the sizeof(int)=4, but the
   sizeof(long)=sizeof(double)=sizeof(size_t)=8. when I checked my
   mpi.h file, I found that the definition about the sizeof(int) is
   correct. meanwhile, I think the mpi.h file was generated according
   to my compute environment when I compiled the Openmpi. So, my codes
   still don't work. :(

   Further, I found when I called the collective routines(such as,
   MPI_Allgatherv(...)) which are implemented by the Point 2 Point
   don't work either when the data > 2GB.

   Thanks
   Xianjun

   2010/12/6 Tim Prince <n8tm@aol.com <mailto:n8tm@aol.com>>

       On 12/5/2010 7:13 PM, Xianjun wrote:

           hi,

           I met a question recently when I tested the MPI_send and
           MPI_Recv
           functions. When I run the following codes, the processes
           hanged and I
           found there was not data transmission in my network at all.

           BTW: I finished this test on two X86-64 computers with 16GB
           memory and
           installed Linux.

           1 #include <stdio.h>
           2 #include <mpi.h>
           3 #include <stdlib.h>
           4 #include <unistd.h>
           5
           6
           7 int main(int argc, char** argv)
           8 {
           9 int localID;
           10 int numOfPros;
           11 size_t Gsize = (size_t)2 * 1024 * 1024 * 1024;
           12
           13 char* g = (char*)malloc(Gsize);
           14
           15 MPI_Init(&argc, &argv);
           16 MPI_Comm_size(MPI_COMM_WORLD, &numOfPros);
           17 MPI_Comm_rank(MPI_COMM_WORLD, &localID);
           18
           19 MPI_Datatype MPI_Type_lkchar;
           20 MPI_Type_contiguous(2048, MPI_BYTE, &MPI_Type_lkchar);
           21 MPI_Type_commit(&MPI_Type_lkchar);
           22
           23 if (localID == 0)
           24 {
           25 MPI_Send(g, 1024*1024, MPI_Type_lkchar, 1, 1,
           MPI_COMM_WORLD);
           26 }
           27
           28 if (localID != 0)
           29 {
           30 MPI_Status status;
           31 MPI_Recv(g, 1024*1024, MPI_Type_lkchar, 0, 1, \
           32 MPI_COMM_WORLD, &status);
           33 }
           34
           35 MPI_Finalize();
           36
           37 return 0;
           38 }

       You supplied all your constants as 32-bit signed data, so, even
       if the count for MPI_Send() and MPI_Recv() were a larger data
       type, you would see this limit. Did you look at your <mpi.h> ?

       --         Tim Prince

       _______________________________________________
       users mailing list

       users@open-mpi.org <mailto:users@open-mpi.org>

       http://www.open-mpi.org/mailman/listinfo.cgi/users



   _______________________________________________
   users mailing list

   users@open-mpi.org <mailto:users@open-mpi.org>

   http://www.open-mpi.org/mailman/listinfo.cgi/users



------------------------------------------------------------------------

_______________________________________________
users mailing list
users@open-mpi.org
http://www.open-mpi.org/mailman/listinfo.cgi/users

_______________________________________________
users mailing list
users@open-mpi.org
http://www.open-mpi.org/mailman/listinfo.cgi/users