[netcdfgroup] Bug in parallel output using nf90_put_var

• To: netcdfgroup@xxxxxxxxxxxxxxxx
• Subject: [netcdfgroup] Bug in parallel output using nf90_put_var
• From: Roman Olson <rolson@xxxxxxxxxxxxxxxxx>
• Date: Mon, 3 Jul 2023 08:57:40 +0900

Dear netCDF community, 

I would like to report a bug in netCDF-Fortran 4.6.0. My Fortran program hangs 
when outputting large arrays in parallel mode on a distributed system. The 
specific function in question is nf90_put_var. The dataset was created in 
parallel mode, and the variable access was set to collective. The variable is 
3D (longitude, latitude, time). When I decrease the time dimension, I can 
output larger longitude/latitude chunks without hanging. When I use a larger 
number of processors, I can output larger datasets. The program works properly 
on a single node (shared memory). In addition, the program also executed 
normally if endianness of the output data is set to “big endian”. 

My netCDF Fortran library has been compiled in parallel mode. My compiler 
information is as follows: ifort (IFORT) 19.0.3.199 20190206. Copyright (C) 
1985-2019 Intel Corporation.  All rights reserved. netCDF information: 
netCDF-Fortran 4.6.0. I am using Fortran coarrays. My sistem information is: 
Linux isotope3 3.10.0-693.el7.x86_64 #1 SMP Tue Aug 22 21:09:27 UTC 2017 x86_64 
x86_64 x86_64 GNU/Linux

Thanks,

Roman Olson

Roman Olson
Researcher, Hydrology Lab
Institute of Industrial Science
University of Tokyo