[Opm] Parallel output in OPM Flow

[Atgeirr Rasmussen]

Dear OPM community,

The current implementation of OPM Flow performs all I/O operations sequentially, also for parallel runs. While this works well for now, it has the potential to become a bottleneck for running large cases in parallel. In the ACROSS project (see https://www.acrossproject.eu for information), we intend to implement parallel output for OPM Flow. The idea is to do this by using the Damaris software library (https://project.inria.fr/damaris/) from INRIA, who are also part of the ACROSS project, and will contribute to the implementation. We think Damaris is a good fit for OPM. It is open source, has been shown to provide good I/O scaling up to large process counts, and has been previously applied in the reservoir simulation domain.

So far, we see three important tasks related to this:
 - Refactoring in OPM Flow to enable the integration of Damaris. From what we have seen so far, any necessary refactoring will be improving the Flow code, and make it more flexible.
 - Integrating the Damaris library. We will ensure that this is an optional dependency, to minimize developer workload and hassle.
 - Agreeing on or creating appropriate file formats for parallel output. Damaris would write to HDF5, but that is quite flexible, so we still have many decisions to make. It is also clear that without the ability to read the files, output is useless. We therefore seek input from all interested stakeholders on this point. Depending on what format we decide on, we may provide an open source stand-alone library for reading the format.

Atgeirr