madness::ExchangeBatchLRU< keyT, dataT > Class Template Reference

Bounded cache of fetched exchange batches, keyed by cloud record key. More...

#include <exchangeoperator.h>

Classes
struct	Slot

Public Member Functions
void	clear ()
	drop every entry; the capacity setting survives
bool	contains (const keyT &key) const
const dataT *	find (const keyT &key)
const dataT &	insert (const keyT &key, dataT &&data, const bool pinned)
	Insert as most-recently-used. pinned marks a batch this rank owns.
std::size_t	n_transient () const
void	set_transient_capacity (const std::size_t c)
	how many non-owned entries may be resident; at least one is always allowed
std::size_t	size () const
std::size_t	transient_capacity () const

Private Attributes
std::list< Slot >	slots_
std::size_t	transient_capacity_ = 2

Detailed Description

template<typename keyT, typename dataT>
class madness::ExchangeBatchLRU< keyT, dataT >

Bounded cache of fetched exchange batches, keyed by cloud record key.

A rank reuses the batches it owns across all of its tasks, so those are pinned and never evicted; only the batches fetched from elsewhere are transient and bounded. An earlier single bound over both let the churning transients evict the reusable owned batches, which re-fetched them at every owned-batch switch. Pinning the owned ones costs a fixed share of the data (orbitals per rank) and is independent of batch granularity.

Entries are held in a list, so a reference returned by find() or insert() stays valid until that entry is evicted — promotion and insertion of other entries do not move it.

Member Function Documentation

clear()

template<typename keyT, typename dataT>

void madness::ExchangeBatchLRU< keyT, dataT >::clear ( )

inline

drop every entry; the capacity setting survives

contains()

template<typename keyT, typename dataT>

bool madness::ExchangeBatchLRU< keyT, dataT >::contains ( const keyT & key ) const

inline

Referenced by test_batch_cache().

find()

template<typename keyT, typename dataT>

const dataT * madness::ExchangeBatchLRU< keyT, dataT >::find ( const keyT & key )

inline

Returns

the cached batch, promoted to most-recently-used, or nullptr if absent

Referenced by test_batch_cache().

insert()

template<typename keyT, typename dataT>

const dataT & madness::ExchangeBatchLRU< keyT, dataT >::insert ( const keyT & key, dataT && data, const bool pinned )

inline

Insert as most-recently-used. pinned marks a batch this rank owns.

Referenced by test_batch_cache().

n_transient()

template<typename keyT, typename dataT>

std::size_t madness::ExchangeBatchLRU< keyT, dataT >::n_transient ( ) const

inline

Referenced by madness::ExchangeBatchLRU< long, std::string >::insert(), and test_batch_cache().

set_transient_capacity()

template<typename keyT, typename dataT>

void madness::ExchangeBatchLRU< keyT, dataT >::set_transient_capacity ( const std::size_t c )

inline

how many non-owned entries may be resident; at least one is always allowed

Referenced by test_batch_cache().

size()

template<typename keyT, typename dataT>

std::size_t madness::ExchangeBatchLRU< keyT, dataT >::size ( ) const

inline

Referenced by test_batch_cache().

transient_capacity()

template<typename keyT, typename dataT>

std::size_t madness::ExchangeBatchLRU< keyT, dataT >::transient_capacity ( ) const

inline

Referenced by test_batch_cache().

Member Data Documentation

slots_

template<typename keyT, typename dataT>

std::list<Slot> madness::ExchangeBatchLRU< keyT, dataT >::slots_

private

transient_capacity_

template<typename keyT, typename dataT>

std::size_t madness::ExchangeBatchLRU< keyT, dataT >::transient_capacity_ = 2

private

---

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

• exchangeoperator.h