Atomistic MD simulations of biomolecules on Fugaku
Fugaku is a supercomputer at RIKEN Center for Computational Science and has 158, 976 nodes. This machine has been made available through the HPCI project.
GENESIS was one of the target applications during the development of this supercomputer and has been developed to maximize the performance of this supercomputer through co-design. The appendix indicate how to execute atomistic MD simulations on Fugaku.
The optimized code for Fugaku is available in GitHub as release verson 2.0.0.
# Download the tutorial file $ git clone https://github.com/genesis-release-r-ccs/genesis.git $ cd genesis $ autoreconf $ ./configure --enable-mixed --host=Fugaku $ make $ make install
To compile GENESIS on Fugaku, please do not forget to add “--host=Fugaku
“.
A option “--enable-single/mixed/double
” controls default precision of real type variables.
We recommend using ”–enable-mixed” in terms of simulation stability and accuracy on Fugaku. In particular, it has been reported that the simulation with “–enable-single” give slightly different box sizes/temperature from those with “–enable-mixed/double”.
You can find further information in doc/GENESIS.pdf
After compile, please execute compile test.
$ cd ./tests/regression_test (make a job script) $ pjsub test_script.sh
Templates of pjsub
script are shown in the Fugaku portal site. (Only registered users) Please find a script with hybrid parallelization type (MPI/OpenMP) in the site. Please find the detailed explanation of pjsub
options from the portal site.
Due to amount of memory, please assign at least 2 nodes for compile test.
./test.py "mpiexec ${bindir}/spdyn " fugaku > regression.log
For normal jobs, please write down the line for the execution of GENESIS.
mpiexec -stdout run_fep1.out ${bindir}/spdyn inp
Although not as fast, GENESIS 1.5.1 and later versions work on Fugaku. If you would like to use the versions, “--host=Fugaku
” should be used for configure, however, only “--enable-single/double
” are available in the versions.
Written by Chigusa Kobayashi@RIKEN Center for Computational Science
July 29, 2022