checkpoint simulator

CHANGELOG.md

@@ -9,6 +9,8 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ### Added
 
+- Add new benchmark to simulation writing a cubed-sphere file using various tunable strategies
+
 ### Changed
 - Modified fpp macro `_UNUSED_DUMMY(x) to use ASSOCIATE instead of PRINT.   With this change it can be used in PURE procedures.
 - Make error handling in Plain_netCDF_Time consistent with MAPL standard error handling

benchmarks/io/CMakeLists.txt

@@ -1,3 +1,4 @@
 add_subdirectory(raw_bw)
 add_subdirectory(gatherv)
 add_subdirectory(combo)
+add_subdirectory(checkpoint_simulator)

benchmarks/io/checkpoint_simulator/CMakeLists.txt

@@ -0,0 +1,14 @@
+set(exe checkpoint_simulator.x)
+
+ecbuild_add_executable (
+  TARGET ${exe}
+  SOURCES checkpoint_simulator.F90 
+  DEFINITIONS USE_MPI)
+
+target_link_libraries (${exe} PRIVATE MAPL.shared MPI::MPI_Fortran FARGPARSE::fargparse esmf )
+target_include_directories (${exe} PUBLIC $<BUILD_INTERFACE:${MAPL_SOURCE_DIR}/include>)
+
+# CMake has an OpenMP issue with NAG Fortran: https://gitlab.kitware.com/cmake/cmake/-/issues/21280
+if (NOT CMAKE_Fortran_COMPILER_ID MATCHES "NAG")
+   target_link_libraries(${exe} PRIVATE OpenMP::OpenMP_Fortran)
+endif ()

benchmarks/io/checkpoint_simulator/README.md

@@ -0,0 +1,19 @@
+This benchmark simulates writing a series of 3D variables of a given cubed-sphere resolution to a file using the same strategies as used by the real checkpoint code in MAPL
+
+The code has the following options and needs an ESMF rc file named checkpoint\_benchmark.rc
+
+- "NX:" the x distribution for each face
+- "NY:" the y distribution for each face
+- "IM\_WORLD:" the cube resolution
+- "LM:" the nubmer of levels
+- "NUM\_WRITERS:" the number of writing processes either to a single or independent files
+- "NUM\_ARRAYS:" the number of 3D variables to write to the file
+- "CHUNK:" whether to chunk, default true
+- "GATHER\_3D:" gather all levels at once (default is false which means a level at a time is gathered)
+- "SPLIT\_FILE:" default false, if true, each writer writes to and independent file
+- "WRITE\_BARRIER:" default false, add a barrier before each write to for synchronization
+- "DO\_WRITES:" default true, if false skips writing (so just an mpi test at that point)
+- "NTRIAL:" default 1, the number of trials to make writing
+- "RANDOM\_DATA:" default true, if true will arrays with random data, if false sets the array to the rank of the process
+
+Note that whatever you set NX and NY to the program must be run on 6*NY*NY processors and the number of writers must evenly divide 6*NY