Cleanup of readers

[HomesGH]

ASCII reader

The method readPhaseSpaceHeader of the ASCII reader is not used in the code anymore. Instead, the data of the config xml was already employed to set according values.

It seems that the method readPhaseSpaceHeader of the ASCII reader is just used for tests, see here.
Additionally, the „final-checkpoint“ is written in ASCII format.

--> We will keep the ASCII reader for now.

MPI_IOCheckpointWriter and MPI_IOReader

@cniethammer @FG-TUM
Do you know if the MPI_IOCheckpointWriter and MPI_IOReader classes are in use or working at all?

I couldn't find any reference and usage of those in the present code base.

On the history of the MPI_IOCheckpointWriter :

• The class was introduced with this commit 4d3fa08 but didn' work correctly
• According to this commit, it was fixed.

[FG-TUM]

After grepping the whole project, it seems MPI_IOCheckpointWriter is not used at all, and MPI_IOReader is only referenced from commented-out code once in PressureGradient, so it is also unused. Both are not registered as plugins, hence, they will probably not work at the moment.

I'd be in favor of removing them!

[cniethammer]

Regarding the MPI_IOCheckpointWriter and MPI_IOReader:
I am not sure if they are working at the moment - but I will try and test.

I think it would be important to have a high-performance I/O plugin in ls1 for checkpointing. MPI functionality can bypass any type-conversions and allows optimal usage of parallel file systems like, e.g., Lustre. So it would be good to keep this.
Especially in the context of WindHPC, this would be important where we might have workflows with short runtimes.

Note: I would also take over the maintainer role for this plugin :)

[HomesGH]

Regarding the MPI_IOCheckpointWriter and MPI_IOReader: I am not sure if they are working at the moment - but I will try and test.

I think it would be important to have a high-performance I/O plugin in ls1 for checkpointing. MPI functionality can bypass any type-conversions and allows optimal usage of parallel file systems like, e.g., Lustre. So it would be good to keep this. Especially in the context of WindHPC, this would be important where we might have workflows with short runtimes.

Note: I would also take over the maintainer role for this plugin :)

I agree that it is important to have a high-performance IO in ls1. However, this should then be the default setting. From a user's perspective of view, you will not dig deep into the different plugins (btw, how many readers/writers are available in ls1? Probably a lot) and just use the default one or the ones from the examples. Those are mainly the binary checkpoint reader/writer and maybe the old ASCII reader/writer (the final-checkpoint option writes in ASCII afaik). So if we want to keep them, we should make them easily usable and set them as default.

[FG-TUM]

I'm in favor of deletion.

@cniethammer what was the result of your tests and do you want to keep this or would you rather reimplement it from scratch?

[cniethammer]

Regarding the MPI_IOCheckpointWriter and the MPI_IOReader:
I had a deeper look into them over the past week. I agree to remove them for now and come up with a new implementation of an MPI-I/O-based writer for fast checkpointing.