On Tue, Sep 21, 2010 at 04:07:01PM -0400, Dan Kokron wrote:
> I am comparing parallel I/O performance between NetCDF-4 and direct
> MPI-IO. I modified the ~netcdf/nc_test4/tst_nc4perf.c to add parameter
> scans for direct MPI-IO calls. It works fine on the discover platform
> using Intel MPI, but fails when using MVAPICH2-1.5.1p1.
>
> Any ideas?
I can tell you what the error message suggests: that you are accessing
more than 2 GB of data in a single call on a platform where an int is
32 bits. True?
==rob
--
Rob Latham
Mathematics and Computer Science Division
Argonne National Lab, IL USA
