Hey guys,

just making you aware of this if you have not already seen it: SPECFEM/specfem3d_globe#531

Any ideas/hints?

Cheers!

Lion

Hej there, I am trying to output specfem 3D globe synthetics in large number, like 100 000 seismograms per component. Writing these to binary sac in parallel is very fast, (for a short, very low frequency test setup about 10 minutes) but the handling of the data becomes extremely cumbersome. Thus, asdf would be perfect for my purposes.

I am running specfem3D_globe on Piz Daint (http://www.cscs.ch/computers/piz_daint_piz_dora/index.html), we managed to compile specfem with asdf, but when running the model with my 100000 receivers the writing stalled. I could watch the size of the synthetics.h5 go up megabyte by megabyte. After two hours I cancelled the job (of which the actual calculation takes a couple of minutes at most). I append the compilation file, Parameters and solver output. I had it write after 1000 time steps, that's why the output stops there. Maybe you can look into this, it would be great if you have an idea what's causing the trouble :) cheers

mk_daint.txt
output_solver.asdf.txt
Par_file.asdf.txt

cmake . -DCMAKE_Fortran_COMPILER=h5pfc
make

[ 85%] Built target write_example_f90
Scanning dependencies of target write_ASDF
[ 92%] Building C object examples/CMakeFiles/write_ASDF.dir/write_ASDF.c.o
[100%] Linking CXX executable write_ASDF
/usr/bin/ld: cannot find -lhdf5
collect2: error: ld returned 1 exit status
make[2]: *** [examples/CMakeFiles/write_ASDF.dir/build.make:92: examples/write_ASDF] Error 1
make[1]: *** [CMakeFiles/Makefile2:181: examples/CMakeFiles/write_ASDF.dir/all] Error 2
make: *** [Makefile:130: all] Error 2

➜ asdf-library git:(master) ✗ h5pfc --version
GNU Fortran (Ubuntu 9.3.0-17ubuntu1~20.04) 9.3.0
Copyright (C) 2019 Free Software Foundation, Inc.
This is free software; see the source for copying conditions. There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

but i have -lhdf5 already.
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/home/ufo/opt/hdf5/lib
export PATH=$PATH:/home/ufo/opt/hdf5/bin
export CPATH=:$CPATH:/home/ufo/opt/hdf5/include

ASDF is now available in SPECFEM3D_GLOBE thanks to @Jas11. There should be a release to make this easier to make a stable installation. Is there a list of items necessarily blocking a release from occurring at this time? Currently, there is one of which I am aware:

• Install pkg-config/cmake files for dependent libraries (which I'm working on.)

line19 in CMakeLists.txt
find_package(HDF5 1.8.0 REQUIRED)
will find the HDF5_LIBRARIES by itself and will not be successful at least on Titan, therefor cause issues like "Could NOT find HDF5 (missing: HDF5_LIBRARIES HDF5_INCLUDE_DIRS) (Required ..."
This can be resolved by safely comment out line 19 and indicating library manually (already done in instruction). Suggested by Matthieu.

Instruction to install the hdf5 parallel library on mac (received from Lion Kirscher)
That might be useful for other so I suggest you add this in the README

Remove everything hdf5 and mpi related from conda

$ conda uninstall hdf5 h5py mpi4py

Install mpi and parallel hdf5 somehow - on mac i strongly recommend

homebrew.

$ brew install open-mpi
$ brew tap homebrew/science
$ brew install hdf5 --with-mpi

Then - with the pip in the conda (environement)

Will make sure mpi4py is linked against your systems MPI

$ pip install mpi4py

Download h5py and unpack it and make sure the conda env is active

$ python setup.py configure --mpi
$ python setup.py build
$ python setup.py install

This will then install a parallel h5py version.

1. rewrite issue

if already exists, it should first clean it out. But it seems not doing so.

2. stationxml issue

stationxml only reaches station level, not channel level

3. trace time issue

* trace starttime and endtime seems wrong
* trace delta and sampling rate wrong

4. quakeml file

 * header information wrong
 * event origin time calculation wrong
 * add more information in mb, Ms

I think I will remove the boost requirement since it is only really used for the parser that is not currently needed. This should make installation somewhat simpler. Let me know if you don't think this is a good idea.

Hi!

I think it would be helpful for users to explicitly state that the HDF5 environment variables have to be set, e.g., HDF5DIR, LD_LIBRARY_PATH addition, etc.

In retrospective it seems obvious, but it would be nice to add as an FYI if Parallel HDF5 is installed manually and not loaded from cluster modules. For the manual case, it would also help the user to know, if HDF5 is installed manually, that it has to be installed with the following flags:

./configure --enable-shared --enable-parallel \
    --enable-fortran --enable-fortran2003 \
    --prefix=<path/to/your/build>

Thanks!

Cheers!