[MPICH] Collective write when only a subset of processes have data to write

[Heshan Lin]

Hmm ... It looks like the problem is related to different MPI
implementations. It had problem with LAM-7.1.2 but worked fine with MPICH2. 

Thanks,
Heshan

> -----Original Message-----
> From: Rajeev Thakur [mailto:thakur at mcs.anl.gov]
> Sent: Monday, February 05, 2007 3:51 PM
> To: 'Heshan Lin'; 'Robert Latham'
> Cc: mpich-discuss at mcs.anl.gov
> Subject: RE: [MPICH] Collective write when only a subset of processes have
data
> to write
> 
> For processes with 0 data, you should be able to call MPI_File_set_view
with
> disp=0, etype=MPI_BYTE, filetype=MPI_BYTE, and MPI_File_write_all with
> count=0. What error do you get if you do this?
> 
> Rajeev
> 
> > -----Original Message-----
> > From: owner-mpich-discuss at mcs.anl.gov
> > [mailto:owner-mpich-discuss at mcs.anl.gov] On Behalf Of Heshan Lin
> > Sent: Monday, February 05, 2007 1:37 PM
> > To: 'Robert Latham'
> > Cc: mpich-discuss at mcs.anl.gov
> > Subject: RE: [MPICH] Collective write when only a subset of
> > processes have data to write
> >
> > Thanks for your response, Rob. But I am not sure how to set
> > parameters for
> > MPI_File_set_view and MPI_File_wirte_all for processes with no data to
> > write. I tried writing 0 length of data but got errors. By
> > any chance you
> > can point me to some examples?
> >
> > Heshan
> >
> > > -----Original Message-----
> > > From: Robert Latham [mailto:robl at mcs.anl.gov]
> > > Sent: Monday, February 05, 2007 10:53 AM
> > > To: Heshan Lin
> > > Cc: mpich-discuss at mcs.anl.gov
> > > Subject: Re: [MPICH] Collective write when only a subset of
> > processes have
> > data
> > > to write
> > >
> > > On Sun, Feb 04, 2007 at 07:59:56PM -0500, Heshan Lin wrote:
> > > > Hi,
> > > >
> > > > I am testing collective write with a parallel program in
> > which every MPI
> > > > process needs to periodically output non-contiguous data.
> > The basic
> > program
> > > > structure for each MPI process looks like following.
> > > >
> > > > MPI_File_open(MPI_COMM_WORLD)
> > > > WHILE (not end) {
> > > >     Computation()
> > > >     MPI_File_set_view()
> > > >     MPI_File_write_all()
> > > > }
> > > >
> > > > One problem I have encountered now is that at some
> > iteration there are
> > only
> > > > a subset of MPI processes having data to write. In that
> > case the program
> > > > will hang if not all processes issue write requests.
> > >
> > > Go ahead and have all processes --  even those with no work to do --
> > > call MPI_File_write_all.  The MPI-IO implementation will
> > know what do
> > > to.
> > >
> > > ==rob
> > >
> > > --
> > > Rob Latham
> > > Mathematics and Computer Science Division    A215 0178 EA2D
> > B059 8CDF
> > > Argonne National Lab, IL USA                 B29D F333 664A
> > 4280 315B
> >
> >