Howdy all!
Supercomputers are expensive devices which turn computation problems into
IO problems.
Too often, many expensive processors are waiting around for IO to complete.
There is a library that can help netCDF users with these issues: the
Parallel IO library (PIO). (https://github.com/NCAR/ParallelIO)
I have been working on this library with the PIO team to add asynchronous
writes and multi-level parallelism with IO running on a small subset of
processors. Distributed arrays functions are provided to assist with
multi-processor programming.
The code is working now, and I am seeking some early adopters to help me
learn how to make it most useful. If you're interested, please contact me.
My intention is to fully integrate the PIO library into the netCDF library,
so that netCDF-based models can easily be converted to take advantage of
these features, with a minimum of code changes to the model.
Keep on netCDFing!
Ed Hartnett
