hype::core::Scheduler Class Reference

The Scheduler is the central component for interaction of the application and the library. More...

#include <scheduler.hpp>

Collaboration diagram for hype::core::Scheduler:

List of all members.

Classes
class	ProcessingDevices
Public Types
typedef std::map< std::string, std::tr1::shared_ptr < Operation > >	MapNameToOperation
Public Member Functions
bool	addAlgorithm (const AlgorithmSpecification &alg_spec, const DeviceSpecification &dev_spec)
	adds an Algorithm to the AlgorithmPool of an operation defined by alg_spec on the processing device defined by dev_spec.
bool	setOptimizationCriterion (const std::string &name_of_operation, const std::string &name_of_optimization_criterion)
	assigns the OptimizationCriterion name_of_optimization_criterion to Operation name_of_operation
bool	setStatisticalMethod (const std::string &name_of_algorithm, const std::string &name_of_statistical_method)
	assigns the StatisticalMethod name_of_statistical_method to an existing Algorithm
bool	setRecomputationHeuristic (const std::string &name_of_algorithm, const std::string &name_of_recomputation_strategy)
	assigns the StatisticalMethod name_of_statistical_method to an existing Algorithm
const SchedulingDecision	getOptimalAlgorithm (const OperatorSpecification &op_spec, const DeviceConstraint &dev_constr)
	Returns a Scheduling Decision, which contains the name of the estimated optimal Algorithm w.r.t. the user specified optimization criterion.
bool	addObservation (const SchedulingDecision &sched_dec, const double &measured_execution_time)
	adds an observed execution time to the algorithm previously choosen by getOptimalAlgorithmName.
const AlgorithmPtr	getAlgorithm (const std::string &name_of_algorithm)
	requests a pointer to Algorithm named name_of_algorithm
EstimatedTime	getEstimatedExecutionTime (const OperatorSpecification &op_spec, const std::string &alg_name)
void	print ()
ProcessingDevices &	getProcessingDevices ()
Static Public Member Functions
static Scheduler &	instance ()
	This method implements the singelton concept for the Scheduler class to avoid multiple instances.
Private Member Functions
	Scheduler ()
	Constructor is private to avoid multiple instances of Scheduler.
	Scheduler (const Scheduler &)
	Copy Constructor is private to avoid copying of the single instance of Scheduler.
Scheduler &	operator= (const Scheduler &)
	Copy assignment operator is private to avoid copying of the single instance of Scheduler.
Private Attributes
MapNameToOperation	map_operationname_to_operation_
	maps the name of an Operation to a pointer to an Operation object
StatisticalMethodMap	map_statisticalmethodname_to_statisticalmethod_
	maps the name of a StatisticalMethod to a pointer to a StatisticalMethod object
ProcessingDevices	proc_devs_
	stores the processing devices, which were specified by the user

---

Detailed Description

The Scheduler is the central component for interaction of the application and the library.

The Scheduler provides two main functionalities. First, it provides the service to decide on the optimal algorithm for an operation w.r.t. a user specified optimization criterion. Second, the Scheduler implements an interface to add new Observations to the executed Algorithm. Hence, it is the central component for interaction of the application and the library. Since it is not meaningful to have multiple instances of the Scheduler class, it is not possible to create multiple Scheduler instances. This property is implemented by using the singelton concept. Additionally, the Scheduler enables the user to setup the Operations with their respective Algorithms as well as to configure for each algorithm a statistical method and a recomputation heuristic and for each operation an optimization criterion. Note that the statistical method and the recomputation statistic can be exchanged at run-time, because the Algortihm uses the pointer to implementation technique (or pimpl-idiom).

Author:

Sebastian Breß

Version:

0.1

Date:

2012

Copyright:

GNU LESSER GENERAL PUBLIC LICENSE - Version 3, http://www.gnu.org/licenses/lgpl-3.0.txt

---

Member Typedef Documentation

typedef std::map<std::string,std::tr1::shared_ptr<Operation> > hype::core::Scheduler::MapNameToOperation

---

Constructor & Destructor Documentation

hype::core::Scheduler::Scheduler

(

)

[private]

Constructor is private to avoid multiple instances of Scheduler.

hype::core::Scheduler::Scheduler

(

const Scheduler &

)

[private]

Copy Constructor is private to avoid copying of the single instance of Scheduler.

Parameters:

reference

to existing Scheduler object

---

Member Function Documentation

bool hype::core::Scheduler::addAlgorithm	(	const AlgorithmSpecification &	alg_spec,
		const DeviceSpecification &	dev_spec
	)

adds an Algorithm to the AlgorithmPool of an operation defined by alg_spec on the processing device defined by dev_spec.

If the specified operation does not exist, it is created. Multiple calls to addAlgorithm with an AlgorithmSpecification having the same Operation name will add the respective algorithms to the algorithm pool of the specified Operation

Parameters:

alg_spec	defines properties of the algorithm, e.g., name, the operation it belongs to, etc.
dev_spec	defines properties of the processing device the algorithm runs on, e.g., device type (CPU or GPU) and the device id

Returns:

returns true on success and false otherwise

bool hype::core::Scheduler::addObservation	(	const SchedulingDecision &	sched_dec,
		const double &	measured_execution_time
	)

adds an observed execution time to the algorithm previously choosen by getOptimalAlgorithmName.

Parameters:

sched_dec	the scheduling decision, this observation belongs to
measured_execution_time	measured execution time, in nanoseconds!!!

Returns:

true on success and false in case an error occured

const AlgorithmPtr hype::core::Scheduler::getAlgorithm

(

const std::string &

name_of_algorithm

)

requests a pointer to Algorithm named name_of_algorithm

Parameters:

name_of_algorithm

name of the Algorithm where the pointer is requested

Returns:

returns a valid pointer to and Algorithm object on success and a NULL pointer if the requested Algorithm was not found

EstimatedTime hype::core::Scheduler::getEstimatedExecutionTime	(	const OperatorSpecification &	op_spec,
		const std::string &	alg_name
	)

const SchedulingDecision hype::core::Scheduler::getOptimalAlgorithm	(	const OperatorSpecification &	op_spec,
		const DeviceConstraint &	dev_constr
	)

Returns a Scheduling Decision, which contains the name of the estimated optimal Algorithm w.r.t. the user specified optimization criterion.

Parameters:

op_spec	OperatorSpecification, contains all available information about the operator to execute
dev_constr	DeviceConstraint, restricting the available algorithms to a subset of the algorithm pool (e.g., allow only CPU algorithms)

Returns:

SchedulingDecision, which contains the suggested algortihm for the specified information

ProcessingDevices& hype::core::Scheduler::getProcessingDevices

(

)

static Scheduler& hype::core::Scheduler::instance

(

)

[static]

This method implements the singelton concept for the Scheduler class to avoid multiple instances.

Returns:

Reference to Scheduler instance.

Scheduler& hype::core::Scheduler::operator=

(

const Scheduler &

)

[private]

Copy assignment operator is private to avoid copying of the single instance of Scheduler.

Parameters:

reference

to existing Scheduler object

void hype::core::Scheduler::print

(

)

bool hype::core::Scheduler::setOptimizationCriterion	(	const std::string &	name_of_operation,
		const std::string &	name_of_optimization_criterion
	)

assigns the OptimizationCriterion name_of_optimization_criterion to Operation name_of_operation

Parameters:

name_of_operation	name of the Operation
name_of_optimization_criterion	Name of OptimizationCriterion

Returns:

returns true on success and false otherwise

bool hype::core::Scheduler::setRecomputationHeuristic	(	const std::string &	name_of_algorithm,
		const std::string &	name_of_recomputation_strategy
	)

assigns the StatisticalMethod name_of_statistical_method to an existing Algorithm

Parameters:

name_of_algorithm	Name of Algorithm
name_of_recomputation_strategy	assigns the RecomputationHeuristic name_of_recomputation_strategy to an existing Algorithm

Returns:

returns true on success and false otherwise

bool hype::core::Scheduler::setStatisticalMethod	(	const std::string &	name_of_algorithm,
		const std::string &	name_of_statistical_method
	)

assigns the StatisticalMethod name_of_statistical_method to an existing Algorithm

Parameters:

name_of_algorithm	Name of Algorithm
name_of_statistical_method	assigns the StatisticalMethod name_of_statistical_method to an existing Algorithm

Returns:

returns true on success and false otherwise

---

Member Data Documentation

MapNameToOperation hype::core::Scheduler::map_operationname_to_operation_ [private]

maps the name of an Operation to a pointer to an Operation object

StatisticalMethodMap hype::core::Scheduler::map_statisticalmethodname_to_statisticalmethod_ [private]

maps the name of a StatisticalMethod to a pointer to a StatisticalMethod object

ProcessingDevices hype::core::Scheduler::proc_devs_ [private]

stores the processing devices, which were specified by the user

---

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

• /home/sebastian/gpudbms/trunk/hype-library/include/core/scheduler.hpp