Hi all,
Recently, I am testing a large domain with EnKF application using JEDI-FV3 v1.1.2.
I have found that analyses from ‘fv3jedi_letkf.x’ are written in ‘double precision’ although background files are in ‘single precision’. Is there any way to change the precision when JEDI writes out analyses files?
(Actually, final I/O takes ~ 20 minutes ( about 20% of walltime) at my applications, So I wonder if I can reduce this more. In addition, my analysis files are written sequentially so I wonder if employing parallel I/O or something could benefit applications.)
Thanks,
Jun
Hi @junpark217,
Have you gotten any assistance with your question? Sorry if no one has helped yet.
I work on the IODA subsystem in JEDI and my current project is to improve the IODA io performance. We are doing exactly what you are suggesting with parallelizing the writing of output files. However, my work is limited to the writing of observation data, not the model related files.
Are you seeing slow io times with model data, observation data or both?
Thanks!
Stephen
Hi Stephen,
Sorry for the late response.
I think the IO times for observation data also seem to be slow … (compared to binary output of GSI)
(maybe model-space is likely to have more file size).
In my personal opinion, I/O time seemed to be better in JEDI-FV3 v1.0.0
Thanks,
Jun