openPMD::MPIBenchmark< DatasetFillerProvider > Class Template Reference

Class representing a benchmark. More...

#include <MPIBenchmark.hpp>

Collaboration diagram for openPMD::MPIBenchmark< DatasetFillerProvider >:

Public Types
using	extentT = Extent::value_type

Public Member Functions
	MPIBenchmark (std::string basePath, Extent tExtent, std::shared_ptr< BlockSlicer > blockSlicer, DatasetFillerProvider dfp, MPI_Comm comm=MPI_COMM_WORLD)
	Construct an MPI benchmark manually. More...
void	addConfiguration (std::string compression, uint8_t compressionLevel, std::string backend, Datatype dt, typename decltype(Series::iterations)::key_type iterations, int threadSize)
void	addConfiguration (std::string compression, uint8_t compressionLevel, std::string backend, Datatype dt, typename decltype(Series::iterations)::key_type iterations)
	Version of addConfiguration() that automatically sets the number of used threads to the MPI size. More...
void	resetConfigurations ()
template<typename Clock >
MPIBenchmarkReport< typename Clock::duration >	runBenchmark (int rootThread=0)
	Main function for running a benchmark. More...

Public Attributes
MPI_Comm	communicator = MPI_COMM_WORLD
Extent	totalExtent
	Total extent of the hypercuboid used in the benchmark.
std::shared_ptr< BlockSlicer >	m_blockSlicer
DatasetFillerProvider	m_dfp

Detailed Description

template<typename DatasetFillerProvider>
class openPMD::MPIBenchmark< DatasetFillerProvider >

Class representing a benchmark.

Allows to configure a benchmark and execute it.

Template Parameters

DatasetFillerProvider

Functor type to create a DatasetFiller with the requested type. Should have a templated operator()() returning a value that can be dynamically casted to a std::shared_ptr<openPMD::DatasetFiller<T>>.

Constructor & Destructor Documentation

MPIBenchmark()

template<typename DatasetFillerProvider >

openPMD::MPIBenchmark< DatasetFillerProvider >::MPIBenchmark	(	std::string	basePath,
		Extent	tExtent,
		std::shared_ptr< BlockSlicer >	blockSlicer,
		DatasetFillerProvider	dfp,
		MPI_Comm	comm = MPI_COMM_WORLD
	)

Construct an MPI benchmark manually.

Parameters

basePath	The path to write to. Will be extended with the backends' filename endings. May be overwritten if performing several benchmarks with the same backend, e.g. when using different compression schemes.
tExtent	The total extent of the dataset.
blockSlicer	An implementation of BlockSlicer class, associating each thread with a portion of the dataset to write to.
dfp	DatasetFillerProvider, a templated functor returning a std::shared_ptr<openPMD::DatasetFiller<T>> or a value dynamically castable to one.
comm	MPI communicator.

Member Function Documentation

addConfiguration() [1/2]

template<typename DatasetFillerProvider >

void openPMD::MPIBenchmark< DatasetFillerProvider >::addConfiguration	(	std::string	compression,
		uint8_t	compressionLevel,
		std::string	backend,
		Datatype	dt,
		typename decltype(Series::iterations)::key_type	iterations
	)

Version of addConfiguration() that automatically sets the number of used threads to the MPI size.

Parameters

compression	Compression string, leave empty to disable commpression.
compressionLevel	Compression level.
backend	Backend to use, specified by filename extension (eg "bp" or "h5").
dt	Type of data to write and read.
iterations	The number of iterations to write and read for each compression strategy. The DatasetFiller functor will be called for each iteration, so it should create sufficient data for one iteration.

addConfiguration() [2/2]

template<typename DatasetFillerProvider >

void openPMD::MPIBenchmark< DatasetFillerProvider >::addConfiguration	(	std::string	compression,
		uint8_t	compressionLevel,
		std::string	backend,
		Datatype	dt,
		typename decltype(Series::iterations)::key_type	iterations,
		int	threadSize
	)

Parameters

compression	Compression string, leave empty to disable commpression.
compressionLevel	Compression level.
backend	Backend to use, specified by filename extension (eg "bp" or "h5").
dt	Type of data to write and read.
iterations	The number of iterations to write and read for each compression strategy. The DatasetFiller functor will be called for each iteration, so it should create sufficient data for one iteration.
threadSize	Number of threads to use.

runBenchmark()

template<typename DatasetFillerProvider >

template<typename Clock >

MPIBenchmarkReport< typename Clock::duration > openPMD::MPIBenchmark< DatasetFillerProvider >::runBenchmark

(

int

rootThread = 0

)

Main function for running a benchmark.

The benchmark is repeated for all previously requested compressions strategies, backends and thread sizes.

Template Parameters

Clock

Clock type to use.

Parameters

rootThread

Rank at which the report will be read.

Returns

A report about the time needed for writing and reading under each compression strategy.

---

The documentation for this class was generated from the following file:

• /tmp/tmp.7iNmvGSbic/openPMD-api/include/openPMD/benchmark/mpi/MPIBenchmark.hpp