Open
Conversation
islice_serial=0 : output field timeslice using parallel hdf5 islice_serial=1 : output field timeslice using serial hdf5 Manual MPI communication + serial hdf5 shows better and stable performance without unexpected super-slow timeslice writing.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The performance of hdf5_write_time_slice is not consistent and can be extremely slow in some cases.
For example, some 3D cases (ITER) can take more than 200 secs to write a time_###.h5 file of about 4 GB.
(see a pdf file attached below for more details)
hdf5_write performance issue.pdf
It seems like the parallel-HDF5 API (with internally implemented MPI communications) somehow causes these inconsistent performance issues.
To resolve the performance issue, this feature allows the user to choose the (manual MPI communication + serial-HDF5 API) as the following:
This new feature is controllable via a new input parameter "islice_serial".
The default value of islice_serial is 0 (islice_serial=0), which uses the original parallel-HDF5 API for outputting the time slice.
If the user set the new input parameter "islice_serial" to 1 (islice_serial=1),
the outputting uses (manual MPI communication + serial-HDF5 API) as described above.
If one uses (islice_serial=1), the performance becomes more consistent and more than 2x faster.
(Tested on Stellar for ITER 3D simulations)
However, there is a caveat on how to truncate the double-precision to single-precision, which should be discussed.