[mpich-discuss] mpiexec hanged up with large stdin

[Pavan Balaji]

Why aren't you reading that data from the application?

I understand that this shouldn't hang in any case.  So, yes, this is a 
bug and we'll fix it (eventually), but Jeff's suggestion to directly 
read from the input file is the correct way to do this.  stdin is not 
the right way to do this, and many MPI implementations don't even 
support it.

  -- Pavan

On 02/28/2012 09:40 AM, Yuheng Xie wrote:
> Jeff,
>
> Thank you for your reply! Although the MPI standard does not require it,
> stdin/stdout and passing data through pipes are traditional conveniences
> in Unix. I'd like MPICH2 improve on these besides performance. So, at
> present it seems I have to generate temporary files before using my data.
>
> --
> Yuheng
>
> 2012/2/28 Jeff Hammond <jhammond at alcf.anl.gov
> <mailto:jhammond at alcf.anl.gov>>
>
>     The MPI standard does not even require stdin to be supported, so it is
>     entirely reasonable for an implementation to support it only up to
>     some reasonable finite value.  If you were to run "cat /dev/random |
>     mpiexec.gforker -n 2 ./test" that program would never return either.
>
>     Since you are intending to use C++, why don't you modify your code to
>     read from an istream associated with a proper input file rather than
>     std::cin?  The source changes should be essentially trivial.
>
>     Jeff
>
>     On Tue, Feb 28, 2012 at 1:15 AM, Yuheng Xie <thinelephant at gmail.com
>     <mailto:thinelephant at gmail.com>> wrote:
>      > Hi,
>      >
>      > I started using MPICH2 recently and found a problem about reading
>     large file
>      > (>64KB) from stdin. My program hanged up on the line MPI_Init(&argc,
>      > &argv);. Could anyone help me? Thanks a lot.
>      >
>      > My program: test.cpp
>      > #include <mpi.h>
>      >
>      > int main(int argc, char **argv)
>      > {
>      >     MPI_Init(&argc, &argv);
>      >     MPI_Finalize();
>      >
>      >     return 0;
>      > }
>      >
>      > Compile and run:
>      > mpicxx test.cpp -o test
>      > cat test.txt | mpiexec.gforker -n 2 ./test   # This is OK
>      > cat test.txt | mpiexec.hydra -n 2 ./test   # When test.txt is
>     larger than
>      > 65536 bytes, this won't return.
>      >
>      > The MPICH2 was installed with following commands:
>      > ./configure --prefix=/usr/local --enable-cxx
>      > --enable-fast=O3,nochkmsg,notiming,ndebug --with-pm=hydra:mpd:gforker
>      > --enable-smpcoll
>      > make
>      > sudo make install
>      >
>      >
>      > _______________________________________________
>      > mpich-discuss mailing list mpich-discuss at mcs.anl.gov
>     <mailto:mpich-discuss at mcs.anl.gov>
>      > To manage subscription options or unsubscribe:
>      > https://lists.mcs.anl.gov/mailman/listinfo/mpich-discuss
>      >
>
>
>
>     --
>     Jeff Hammond
>     Argonne Leadership Computing Facility
>     University of Chicago Computation Institute
>     jhammond at alcf.anl.gov <mailto:jhammond at alcf.anl.gov> / (630) 252-5381
>     http://www.linkedin.com/in/jeffhammond
>     https://wiki.alcf.anl.gov/old/index.php/User:Jhammond
>     https://wiki-old.alcf.anl.gov/index.php/User:Jhammond
>     _______________________________________________
>     mpich-discuss mailing list mpich-discuss at mcs.anl.gov
>     <mailto:mpich-discuss at mcs.anl.gov>
>     To manage subscription options or unsubscribe:
>     https://lists.mcs.anl.gov/mailman/listinfo/mpich-discuss
>
>
>
>
> _______________________________________________
> mpich-discuss mailing list     mpich-discuss at mcs.anl.gov
> To manage subscription options or unsubscribe:
> https://lists.mcs.anl.gov/mailman/listinfo/mpich-discuss

-- 
Pavan Balaji
http://www.mcs.anl.gov/~balaji