Peano

Searching...

No Matches

#include <AbstractLoadBalancing.h>

Inheritance diagram for toolbox::loadbalancing::AbstractLoadBalancing:

[legend]

Collaboration diagram for toolbox::loadbalancing::AbstractLoadBalancing:

[legend]

Public Member Functions
	AbstractLoadBalancing (Configuration configuration, CostMetrics costMetrics)
	Constructor.

virtual	~AbstractLoadBalancing ()
	Destructor.

virtual std::string	toString () const
	Create a string representation of load balancing object.

virtual void	enable (bool)
	Switch on/off.

virtual bool	hasSplitRecently () const
	Has mesh split recently.

bool	isEnabled (bool globally) const
	Is load balancing enabled.

virtual int	getGlobalNumberOfTrees () const
	Delegate to stats.

virtual bool	hasStagnated () const
	A load balancing can either be stagnating or be switched off for this predicate to hold.

virtual void	finishSimulation ()

virtual void	finishStep ()=0
	Finish the step.

void	setConfigurationAndMetricsNullWithoutDelete ()
	This is only used when you concatenate balancing rules and you want to disable any deletion.

Static Public Attributes
static constexpr int	NoHeaviestTreeAvailable = -1
	Is used by tree identification and either indicates that there are no trees at all or means that the heaviest tree is on the blacklist.

Protected Member Functions
bool	fitsIntoMemory (State state) const
	Ensure enough memory is left-over.

bool	isInterRankBalancingBad () const
	Is the balancing between the ranks ok.

bool	isIntraRankBalancingBad () const
	Is the balancing on the rank ok.

bool	areRanksUnemployed () const

double	getWeightOfHeaviestLocalSpacetree () const

int	getIdOfHeaviestLocalSpacetree () const
	Determines the maximum spacetree size a tree should have in the optimal case.

int	getIdOfHeaviestLocalSpacetree (double tolerance) const
	Similar to getIdOfHeaviestLocalSpacetree() but you might get one of the trees back that is close to the heaviest one up to tolerance.

Protected Attributes
Blacklist	_blacklist

Statistics	_statistics

Configuration *	_configuration

CostMetrics *	_costMetrics

State	_state
	Ensure that you invoke.

Static Protected Attributes
static tarch::logging::Log	_log

Detailed Description

Definition at line 24 of file AbstractLoadBalancing.h.

Constructor & Destructor Documentation

◆ AbstractLoadBalancing()

toolbox::loadbalancing::AbstractLoadBalancing::AbstractLoadBalancing	(	Configuration *	configuration,
		CostMetrics *	costMetrics )

Constructor.

Data ownership is handed over to the load balancing, i.e. the balancing has to delete the configuration object unless you pass in the nullptr. If you pass in the nullptr, you cannot use the object, so this should be done with care by any subclass.

Many implementations of AbstractLoadBalancing change the _state field immediately in their constructor. Please ensure that you invoke

  _statistics.notifyOfStateChange(_state);

afterwards to ensure that the statistics are correct right from the start.

Rationale

I originally thought it might be reasonable to add assertions to ensure that configuration and costMetrics are not nullptr. However, there are specific hard-coded load balancing schemes which do explicitly use nullptr arguments. Hardcoded is the prime example. Therefore, I do not nullptr checks here and I leave it to subclasses to add the corresponding assertions if they require proper pointers here.

◆ ~AbstractLoadBalancing()

virtual toolbox::loadbalancing::AbstractLoadBalancing::~AbstractLoadBalancing ( )

virtual

Destructor.

Free the configuration and cost metrics objects. See the discussion in the constructor for contextual information re the validity of nullptr arguments here.

Member Function Documentation

◆ areRanksUnemployed()

bool toolbox::loadbalancing::AbstractLoadBalancing::areRanksUnemployed ( ) const

protected

◆ enable()

virtual void toolbox::loadbalancing::AbstractLoadBalancing::enable ( bool )

virtual

Switch on/off.

Reimplemented in toolbox::loadbalancing::strategies::cascade::Cascade< HostedLoadBalancing0, HostedLoadBalancing1 >, and toolbox::loadbalancing::strategies::NoLoadBalancing.

Referenced by toolbox::loadbalancing::strategies::cascade::Cascade< HostedLoadBalancing0, HostedLoadBalancing1 >::enable().

Here is the caller graph for this function:

◆ finishSimulation()

virtual void toolbox::loadbalancing::AbstractLoadBalancing::finishSimulation ( )

virtual

Reimplemented in toolbox::loadbalancing::strategies::cascade::Cascade< HostedLoadBalancing0, HostedLoadBalancing1 >, and toolbox::loadbalancing::strategies::Hardcoded.

◆ finishStep()

virtual void toolbox::loadbalancing::AbstractLoadBalancing::finishStep ( )

pure virtual

Finish the step.

No matter what you do here, you have to invoke

     _statistics.updateGlobalView();
     _blacklist.update();

in this routine.

◆ fitsIntoMemory()

bool toolbox::loadbalancing::AbstractLoadBalancing::fitsIntoMemory ( State state ) const

protected

Ensure enough memory is left-over.

◆ getGlobalNumberOfTrees()

virtual int toolbox::loadbalancing::AbstractLoadBalancing::getGlobalNumberOfTrees ( ) const

virtual

Delegate to stats.

Reimplemented in toolbox::loadbalancing::strategies::cascade::Cascade< HostedLoadBalancing0, HostedLoadBalancing1 >.

◆ getIdOfHeaviestLocalSpacetree() [1/2]

int toolbox::loadbalancing::AbstractLoadBalancing::getIdOfHeaviestLocalSpacetree ( ) const

protected

Determines the maximum spacetree size a tree should have in the optimal case.

As this routine does not really adopt the blacklist, it can introduce cyclles: If we repeatedly try to split the same rank this means that we have tried to split it, have not been successful, and tried again. This can happen, as not all trees can be split. See peano4::grid::Spacetree::isCellTopDownSplitCandidate() and peano4::grid::Spacetree::isCellBottomUpSplitCandidate() for a discussion which cells can be split and which can't. As not all cells can't be given away, not all trees can be split up.

Randomisation

If two or more trees have the same weight, the routine should randomly return one of them.

Parameters

tolerance By default, you get the last heaviest tree. But you can ask for a random tree

Returns: NoHeaviestTreeAvailable If there are no local trees or if the heaviest tree is on the blacklist, i.e. we have to assume that it still is splitting.

◆ getIdOfHeaviestLocalSpacetree() [2/2]

int toolbox::loadbalancing::AbstractLoadBalancing::getIdOfHeaviestLocalSpacetree ( double tolerance ) const

protected

Similar to getIdOfHeaviestLocalSpacetree() but you might get one of the trees back that is close to the heaviest one up to tolerance.

Tolerance here is the relative difference: So you are guaranteed that

\( m_i \geq tolerance \cdot m_{max} \)

\( \frac{m_i}{m_{max}} \geq tolerance \)

◆ getWeightOfHeaviestLocalSpacetree()

double toolbox::loadbalancing::AbstractLoadBalancing::getWeightOfHeaviestLocalSpacetree ( ) const

protected

Returns: -1 if there is no local tree yet

◆ hasSplitRecently()

virtual bool toolbox::loadbalancing::AbstractLoadBalancing::hasSplitRecently ( ) const

virtual

Has mesh split recently.

Means that mesh has split over the past three iterations. The number three is hard-coded, as we need three iterations to perform a proper split:

Trigger the split,
split the mesh actually,
update all the stats.

If we have then not come to the conclusion that we should immediately split again, then we haven't split recently.

Reimplemented in toolbox::loadbalancing::strategies::cascade::Cascade< HostedLoadBalancing0, HostedLoadBalancing1 >.

◆ hasStagnated()

virtual bool toolbox::loadbalancing::AbstractLoadBalancing::hasStagnated ( ) const

virtual

A load balancing can either be stagnating or be switched off for this predicate to hold.

Reimplemented in toolbox::loadbalancing::strategies::cascade::Cascade< HostedLoadBalancing0, HostedLoadBalancing1 >.

◆ isEnabled()

bool toolbox::loadbalancing::AbstractLoadBalancing::isEnabled ( bool globally ) const

Is load balancing enabled.

Clarifies whether load balancing is, in principle, enabled. We we return true, it means that we are not disabled. The balancer might still be temporarily suspended.

◆ isInterRankBalancingBad()

bool toolbox::loadbalancing::AbstractLoadBalancing::isInterRankBalancingBad ( ) const

protected

Is the balancing between the ranks ok.

This operation checks if the inter-rank load decomposition does violate the overall load balancing constraints. The global balancing is determined by a cascade of checks:

If we have an inconsistent data view, i.e. if the stats are not valid, we have to assume that everything is fine.
If we have fewer trees globally than we have ranks, the balancing is by definition poor.
If we host more trees than allowed, the balancing by definition is good, since we could not refine further anyway.
Otherwise, we look if the max load per rank and the min load per rank differ more than the tolerance.

◆ isIntraRankBalancingBad()

bool toolbox::loadbalancing::AbstractLoadBalancing::isIntraRankBalancingBad ( ) const

protected

Is the balancing on the rank ok.

Similar to isInterRankBalancingBad(), this operation runs through a series of checks:

If there are no trees on the rank, then the result is false. By definition, everything is well-balanced.
If there is only one tree, the balancing is by definition bad, as we cannot make a balancing statement.
If we host more trees than allowed, the balancing by definition is good, since we could not refine further anyway.
Otherwise, we compare the minimum and maximum tree.

◆ setConfigurationAndMetricsNullWithoutDelete()

void toolbox::loadbalancing::AbstractLoadBalancing::setConfigurationAndMetricsNullWithoutDelete ( )

This is only used when you concatenate balancing rules and you want to disable any deletion.

Please do not invoke it unless you know exactly what you are doing.

◆ toString()

virtual std::string toolbox::loadbalancing::AbstractLoadBalancing::toString ( ) const

virtual

Create a string representation of load balancing object.

Generic string serialisation. You might want to extend it. Therefore, the routine is labelled as virtual.

The routine has to be able to deal with the fact that _costMetrics might be nullptr according to AbstractLoadBalancing(). The configuration in return does not feed into toString(). Therefore, we don't have to check if it is pointing to nullptr.

Reimplemented in toolbox::loadbalancing::strategies::RecursiveBipartition, and toolbox::loadbalancing::strategies::SplitOversizedTree.

Field Documentation

◆ _blacklist

Blacklist toolbox::loadbalancing::AbstractLoadBalancing::_blacklist

protected

Definition at line 155 of file AbstractLoadBalancing.h.

◆ _configuration

Configuration* toolbox::loadbalancing::AbstractLoadBalancing::_configuration

protected

Definition at line 159 of file AbstractLoadBalancing.h.

◆ _costMetrics

CostMetrics* toolbox::loadbalancing::AbstractLoadBalancing::_costMetrics

protected

Definition at line 161 of file AbstractLoadBalancing.h.

◆ _log

tarch::logging::Log toolbox::loadbalancing::AbstractLoadBalancing::_log

staticprotected

Definition at line 153 of file AbstractLoadBalancing.h.

◆ _state

State toolbox::loadbalancing::AbstractLoadBalancing::_state

protected

Ensure that you invoke.

   _statistics.notifyOfStateChange(_state)

whenever you change the state. Otherwise, the MPI will have an invalid view of the whole world.

Definition at line 171 of file AbstractLoadBalancing.h.

Referenced by toolbox::loadbalancing::strategies::cascade::Cascade< HostedLoadBalancing0, HostedLoadBalancing1 >::hasStagnated().

◆ _statistics

Statistics toolbox::loadbalancing::AbstractLoadBalancing::_statistics

protected

Definition at line 157 of file AbstractLoadBalancing.h.

◆ NoHeaviestTreeAvailable

constexpr int toolbox::loadbalancing::AbstractLoadBalancing::NoHeaviestTreeAvailable = -1

staticconstexpr

Is used by tree identification and either indicates that there are no trees at all or means that the heaviest tree is on the blacklist.

See implementation remarks in class description.

Definition at line 31 of file AbstractLoadBalancing.h.

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

src/toolbox/loadbalancing/AbstractLoadBalancing.h

Public Member Functions

Static Public Attributes

Protected Member Functions

Protected Attributes

Static Protected Attributes

Detailed Description

Constructor & Destructor Documentation

◆ AbstractLoadBalancing()

Rationale

◆ ~AbstractLoadBalancing()

Member Function Documentation

◆ areRanksUnemployed()

◆ enable()

◆ finishSimulation()

◆ finishStep()

◆ fitsIntoMemory()

◆ getGlobalNumberOfTrees()

◆ getIdOfHeaviestLocalSpacetree() [1/2]

Randomisation

◆ getIdOfHeaviestLocalSpacetree() [2/2]

◆ getWeightOfHeaviestLocalSpacetree()

◆ hasSplitRecently()

◆ hasStagnated()

◆ isEnabled()

◆ isInterRankBalancingBad()

◆ isIntraRankBalancingBad()

◆ setConfigurationAndMetricsNullWithoutDelete()

◆ toString()

Field Documentation

◆ _blacklist

◆ _configuration

◆ _costMetrics

◆ _log

◆ _state

◆ _statistics

◆ NoHeaviestTreeAvailable