ck-core/ckcausalmlog.h File Reference

Go to the source code of this file.

Data Structures

struct Determinant

Struct to store the determinant of a particular message. More...

struct RemoveDeterminantsHeader

Struct for the header of the removeDeterminants handler. More...

struct StoreDeterminantsHeader

Struct for the header of the storeDeterminants handler. More...

class Ticket

Structure for a ticket assigned to a particular message. More...

class SNToTicket

Class that maps SN (sequence numbers) to TN (ticket numbers) for a particular object. More...

class ChareMlogData

This file includes the definition of the class for storing the meta data associdated with the message logging protocol. More...

class MlogEntry

Entry in a message log. More...

class LocationID

Class that represents the location of an array element. More...

class StoredCheckpoint

class RestoredLocalMap

Class for storing metadata of local messages. More...

struct TicketRequest

struct TicketReply

struct BufferedLocalLogHeader

struct CheckPointDataMsg

struct TProcessedLog

struct RestartRequest

Struct to request a particular action during restart. More...

struct RestartProcessorData

struct ResendRequest

struct ReceivedTNData

struct ReceivedDetData

struct ResendData

struct MigrationRecord

struct MigrationNotice

struct MigrationNoticeAck

struct RetainedMigratedObject

struct VerifyAckMsg

struct CheckpointBarrierMsg

struct CurrentLocationMsg

struct LBStepMsg

struct DummyMigrationMsg

Typedefs

typedef CkHashtableT
< CkHashtableAdaptorT< CkObjID >
, CkVec< Determinant > * > CkDeterminantHashtableT

Typedef for the hashtable type that maps object IDs to determinants.

typedef BufferedLocalLogHeader BufferedTicketRequestHeader

typedef CheckPointDataMsg CheckPointAck

typedef RestartRequest CkPingMsg

typedef RestartRequest CheckpointRequest

typedef ResendRequest RemoveLogRequest

typedef void(* MlogFn )(void *, ChareMlogData *)

Functions

CpvExtern (Chare *, _currentObj)

PUPbytes (Ticket) class MlogEntry

CpvExtern (CkQ< TicketRequest * > *, _delayedTicketRequests)

CpvExtern (CkQ< MlogEntry * > *, _delayedLocalTicketRequests)

CpvExtern (char **, _bufferedTicketRequests)

void _messageLoggingInit ()

Initialize message logging data structures and register handlers.

void sendGroupMsg (envelope *env, int destPE, int _infoIdx)

Sends a group message that might be a broadcast.

void sendArrayMsg (envelope *env, int destPE, int _infoIdx)

Sends a message to an array element.

void sendChareMsg (envelope *env, int destPE, int _infoIdx, const CkChareID *pCid)

Sends a message to a singleton chare.

void sendNodeGroupMsg (envelope *env, int destNode, int _infoIdx)

Sends a nodegroup message that might be a broadcast.

void sendCommonMsg (CkObjID &recver, envelope *env, int destPE, int _infoIdx)

A method to generate the actual ticket requests for groups, nodegroups or arrays.

void sendMsg (CkObjID &sender, CkObjID &recver, int destPE, MlogEntry *entry, MCount SN, MCount TN, int resend)

Method that does the actual send by creating a ticket request filling it up and sending it.

void sendLocalMsg (MlogEntry *entry)

Function to send a local message.

void _ticketRequestHandler (TicketRequest *)

If there are any delayed requests, process them first before processing this request.

void _ticketHandler (TicketReply *)

This function handles the ticket received after a request.

void _pingHandler (CkPingMsg *msg)

void _bufferedLocalMessageCopyHandler (BufferedLocalLogHeader *recvdHeader, int freeHeader=1)

void _bufferedLocalMessageAckHandler (BufferedLocalLogHeader *recvdHeader)

void _bufferedTicketRequestHandler (BufferedTicketRequestHeader *recvdHeader)

Handler used for dealing with a bunch of ticket requests from one processor.

void _bufferedTicketHandler (BufferedTicketRequestHeader *recvdHeader)

Message to handle the bunch of tickets that we get from one processor.

void _storeDeterminantsHandler (char *buffer)

Stores the determinants coming from other processor.

void _removeDeterminantsHandler (char *buffer)

Removes the determinants after a particular index in the _localDets array.

void _skipCldEnqueue (int pe, envelope *env, int infoFn)

void _noCldNodeEnqueue (int node, envelope *env)

void generalCldEnqueue (int destPE, envelope *env, int _infoIdx)

void retryTicketRequest (void *_ticketRequest, double curWallTime)

int preProcessReceivedMessage (envelope *env, Chare **objPointer, MlogEntry **localLogEntry)

void postProcessReceivedMessage (Chare *obj, CkObjID &sender, MCount SN, MlogEntry *entry)

Updates a few variables once a message has been processed.

CpvExtern (StoredCheckpoint *, _storedCheckpointData)

void checkpointAlarm (void *_dummy, double curWallTime)

void startMlogCheckpoint (void *_dummy, double curWallTime)

Starts the checkpoint phase after migration.

void pupArrayElementsSkip (PUP::er &p, CmiBool create, MigrationRecord *listToSkip, int listSize=0)

Pups all the array elements in this processor.

void _checkpointRequestHandler (CheckpointRequest *request)

void _storeCheckpointHandler (char *msg)

void _checkpointAckHandler (CheckPointAck *ackMsg)

void _removeProcessedLogHandler (char *requestMsg)

Removes messages in the log according to the received ticket numbers.

void garbageCollectMlog ()

Garbage collects the message log and other data structures.

void CkMlogRestart (const char *dummy, CkArgMsg *dummyMsg)

Function for restarting the crashed processor.

void CkMlogRestartDouble (void *, double)

void processReceivedTN (Chare *obj, int vecsize, MCount *listTNs)

Processes the received list of tickets from a particular PE.

void processReceivedDet (Chare *obj, int vecsize, Determinant *listDets)

Processes the received list of determinants from a particular PE.

void initializeRestart (void *data, ChareMlogData *mlogData)

Initializes variables and flags for restarting procedure.

void distributeRestartedObjects ()

Distributes objects to accelerate recovery after a failure.

void sendDummyMigration (int restartPE, CkGroupID lbID, CkGroupID locMgrID, CkArrayIndexMax &idx, int locationPE)

this method is used to send messages to a restarted processor to tell it that a particular expected object is not going to get to it

void CkMlogRestartLocal ()

void _getCheckpointHandler (RestartRequest *restartMsg)

Gets the stored checkpoint for its buddy processor.

void _recvCheckpointHandler (char *_restartData)

Receives the checkpoint data from its buddy, restores the state of all the objects and asks everyone else to update its home.

void _resendMessagesHandler (char *msg)

Resends messages since last checkpoint to the list of objects included in the request.

void _sendDetsHandler (char *msg)

Send all remote determinants to a particular failed PE.

void _sendDetsReplyHandler (char *msg)

Receives determinants stored on remote nodes.

void _receivedTNDataHandler (ReceivedTNData *msg)

Receives a list of TNs coming from the home PE of a migrated object (parallel restart).

void _receivedDetDataHandler (ReceivedDetData *msg)

Receives a list of determinants coming from the home PE of a migrated object (parallel restart).

void _distributedLocationHandler (char *receivedMsg)

Handler to update information about an object just received.

void _updateHomeRequestHandler (RestartRequest *updateRequest)

Updates the homePe for all chares in this processor.

void _updateHomeAckHandler (RestartRequest *updateHomeAck)

Receives the updateHome ACKs from all other processors.

void _verifyAckRequestHandler (VerifyAckMsg *verifyRequest)

void _verifyAckHandler (VerifyAckMsg *verifyReply)

void _dummyMigrationHandler (DummyMigrationMsg *msg)

this handler is used to process a dummy migration msg.

void _restartHandler (RestartRequest *restartMsg)

Function to restart this processor.

void _getRestartCheckpointHandler (RestartRequest *restartMsg)

Gets the stored checkpoint but calls another function in the sender.

void _recvRestartCheckpointHandler (char *_restartData)

Receives the checkpoint coming from its buddy.

void startLoadBalancingMlog (void(*fnPtr)(void *), void *_centralLb)

Load Balancing.

void finishedCheckpointLoadBalancing ()

void sendMlogLocation (int targetPE, envelope *env)

void resumeFromSyncRestart (void *data, ChareMlogData *mlogData)

void _receiveMlogLocationHandler (void *buf)

void _receiveMigrationNoticeHandler (MigrationNotice *msg)

void _receiveMigrationNoticeAckHandler (MigrationNoticeAck *msg)

void _getGlobalStepHandler (LBStepMsg *msg)

void _recvGlobalStepHandler (LBStepMsg *msg)

Receives the global step handler from PE 0.

void _checkpointBarrierHandler (CheckpointBarrierMsg *msg)

Processor 0 receives a contribution from every other processor after checkpoint.

void _checkpointBarrierAckHandler (CheckpointBarrierMsg *msg)

int getCheckPointPE ()

Getting the pe number of the current processor's buddy.

int isSameDet (Determinant *first, Determinant *second)

void forAllCharesDo (MlogFn fnPointer, void *data)

Map function pointed by fnPointer over all the chares living in this processor.

envelope * copyEnvelope (envelope *env)

void _initDone (void)

This function (not a handler) is called once and only once per processor.

void _resetNodeBocInitVec (void)

void informLocationHome (CkGroupID mgrID, CkArrayIndexMax idx, int homePE, int currentPE)

method that informs an array elements home processor of its current location It is a converse method to bypass the charm++ message logging framework

void _receiveLocationHandler (CurrentLocationMsg *data)

void CmiDeliverRemoteMsgHandlerRange (int lowerHandler, int higherHandler)

Variables

char objString [100]

int _maxBufferedTicketRequests

int _checkpointRequestHandlerIdx

int _storeCheckpointHandlerIdx

int _checkpointAckHandlerIdx

int _removeProcessedLogHandlerIdx

int _getCheckpointHandlerIdx

int _recvCheckpointHandlerIdx

int _resendMessagesHandlerIdx

int _sendDetsHandlerIdx

int _sendDetsReplyHandlerIdx

int _receivedTNDataHandlerIdx

int _receivedDetDataHandlerIdx

int _distributedLocationHandlerIdx

int _updateHomeRequestHandlerIdx

int _updateHomeAckHandlerIdx

int _verifyAckRequestHandlerIdx

int _verifyAckHandlerIdx

int _dummyMigrationHandlerIdx

int onGoingLoadBalancing

For testing on clusters we might carry out restarts on a porcessor without actually starting it 1 -> false restart 0 -> restart after an actual crash.

void * centralLb

void(* resumeLbFnPtr )(void *)

int _receiveMlogLocationHandlerIdx

int _receiveMigrationNoticeHandlerIdx

int _receiveMigrationNoticeAckHandlerIdx

int _getGlobalStepHandlerIdx

int _recvGlobalStepHandlerIdx

int _checkpointBarrierHandlerIdx

int _checkpointBarrierAckHandlerIdx

CkVec< MigrationRecord > migratedNoticeList

CkVec< RetainedMigratedObject * > retainedObjectList

int _receiveLocationHandlerIdx

Typedef Documentation

typedef CkHashtableT<CkHashtableAdaptorT<CkObjID>, CkVec<Determinant> *> CkDeterminantHashtableT

Typedef for the hashtable type that maps object IDs to determinants.

Definition at line 53 of file ckcausalmlog.h.

typedef BufferedLocalLogHeader BufferedTicketRequestHeader

Definition at line 360 of file ckcausalmlog.h.

typedef CheckPointDataMsg CheckPointAck

Definition at line 374 of file ckcausalmlog.h.

typedef RestartRequest CkPingMsg

Definition at line 390 of file ckcausalmlog.h.

typedef RestartRequest CheckpointRequest

Definition at line 391 of file ckcausalmlog.h.

typedef ResendRequest RemoveLogRequest

Definition at line 411 of file ckcausalmlog.h.

typedef void(* MlogFn)(void *, ChareMlogData *)

Definition at line 507 of file ckcausalmlog.h.

Function Documentation

CpvExtern	(	Chare *	,
		_currentObj
	)

PUPbytes ( Ticket )

CpvExtern	(	CkQ< TicketRequest * > *	,
		_delayedTicketRequests
	)

CpvExtern	(	CkQ< MlogEntry * > *	,
		_delayedLocalTicketRequests
	)

CpvExtern	(	char **	,
		_bufferedTicketRequests
	)

void _messageLoggingInit ( )

Initialize message logging data structures and register handlers.

Definition at line 237 of file ckcausalmlog.C.

void sendGroupMsg	(	envelope *	env,
		int	destPE,
		int	_infoIdx
	)

Sends a group message that might be a broadcast.

Definition at line 469 of file ckcausalmlog.C.

References CkCopyMsg(), Converse::CkMyPe(), CmiMyPe(), CkObjID::data, EnvToUsr(), envelope::getGroupNum(), _ObjectID::group, _ObjectID::id, _ObjectID::onPE, printf(), sendCommonMsg(), envelope::sender, sendGroupMsg(), envelope::SN, envelope::TN, CkObjID::type, TypeGroup, TypeInvalid, and UsrToEnv().

Referenced by _sendMsgBranch(), and sendGroupMsg().

void sendArrayMsg	(	envelope *	env,
		int	destPE,
		int	_infoIdx
	)

Sends a message to an array element.

Definition at line 534 of file ckcausalmlog.C.

References _ObjectID::array, CkArrayIndexBase::asChild(), Converse::CkMyPe(), CkObjID::data, envelope::getsetArrayIndex(), envelope::getsetArrayMgr(), _ObjectID::s_array::id, _ObjectID::s_array::idx, printf(), sendCommonMsg(), envelope::SN, envelope::TN, CkObjID::toString(), CkObjID::type, and TypeArray.

Referenced by CkArrayManagerDeliver().

void sendChareMsg	(	envelope *	env,
		int	destPE,
		int	_infoIdx,
		const CkChareID *	pCid
	)

Sends a message to a singleton chare.

Definition at line 556 of file ckcausalmlog.C.

References _ObjectID::chare, Converse::CkMyPe(), CkObjID::data, _ObjectID::id, printf(), sendCommonMsg(), envelope::SN, envelope::TN, CkObjID::toString(), CkObjID::type, TypeArray, and TypeChare.

Referenced by CkSendMsg().

void sendNodeGroupMsg	(	envelope *	env,
		int	destNode,
		int	_infoIdx
	)

Sends a nodegroup message that might be a broadcast.

Definition at line 502 of file ckcausalmlog.C.

References CkCopyMsg(), Converse::CkMyPe(), CkObjID::data, EnvToUsr(), envelope::getGroupNum(), _ObjectID::group, _ObjectID::id, _ObjectID::onPE, printf(), sendCommonMsg(), envelope::sender, sendNodeGroupMsg(), envelope::SN, envelope::TN, CkObjID::type, TypeInvalid, TypeNodeGroup, and UsrToEnv().

Referenced by _sendMsgNodeBranch(), and sendNodeGroupMsg().

void sendCommonMsg	(	CkObjID &	recver,
		envelope *	env,
		int	destPE,
		int	_infoIdx
	)

A method to generate the actual ticket requests for groups, nodegroups or arrays.

Definition at line 577 of file ckcausalmlog.C.

References _startTime, Converse::CkMyPe(), CmiMemoryCheck(), copyEnvelope(), generalCldEnqueue(), SNToTicket::get(), CkHashtableT< KEY, OBJ >::get(), CkObjID::getObject(), envelope::incarnation, isLocal(), isTeamLocal(), Chare::mlogData, ChareMlogData::nextSN(), printf(), envelope::recver, envelope::sender, sendLocalMsg(), sendMsg(), envelope::SN, ChareMlogData::teamTable, Ticket::TN, CkObjID::toString(), CkObjID::type, and TypeInvalid.

Referenced by sendArrayMsg(), sendChareMsg(), sendGroupMsg(), and sendNodeGroupMsg().

void sendMsg	(	CkObjID &	sender,
		CkObjID &	recver,
		int	destPE,
		MlogEntry *	entry,
		MCount	SN,
		MCount	TN,
		int	resend
	)

Method that does the actual send by creating a ticket request filling it up and sending it.

Definition at line 684 of file ckcausalmlog.C.

References MlogEntry::_infoIdx, ChareMlogData::addLogEntry(), Converse::CkMyPe(), CmiMemoryCheck(), CmiMyPe(), CmiSyncVectorSend(), MlogEntry::destPE, MlogEntry::env, envelope::freeMsg, generalCldEnqueue(), CkObjID::getObject(), envelope::getTotalsize(), StoreDeterminantsHeader::index, MlogEntry::indexBufDets, isTeamLocal(), Chare::mlogData, MLOGFT_totalLogSize, MLOGFT_totalMessages, StoreDeterminantsHeader::number, MlogEntry::numBufDets, StoreDeterminantsHeader::PE, StoreDeterminantsHeader::phase, printf(), envelope::recver, envelope::sender, sizes, envelope::SN, envelope::TN, CkObjID::toString(), and totalMsgsSize.

Referenced by sendCommonMsg().

void sendLocalMsg ( MlogEntry * entry )

Function to send a local message.

It first gets a ticket and then enqueues the message. If we are recovering, then the message is enqueued in a delay queue.

Definition at line 765 of file ckcausalmlog.C.

References MlogEntry::_infoIdx, _skipCldEnqueue(), _startTime, CmiMemoryCheck(), CmiMyPe(), MlogEntry::env, CkObjID::getObject(), printf(), envelope::recver, envelope::sender, envelope::SN, and CkObjID::toString().

Referenced by sendCommonMsg().

void _ticketRequestHandler ( TicketRequest * ) [inline]

If there are any delayed requests, process them first before processing this request.

Definition at line 928 of file ckcausalmlog.C.

void _ticketHandler ( TicketReply * ) [inline]

This function handles the ticket received after a request.

Definition at line 965 of file ckmessagelogging.C.

Referenced by _bufferedTicketHandler(), and _messageLoggingInit().

void _pingHandler ( CkPingMsg * msg )

Definition at line 1164 of file ckcausalmlog.C.

void _bufferedLocalMessageCopyHandler	(	BufferedLocalLogHeader *	recvdHeader,
		int	freeHeader = `1`
	)

Definition at line 1110 of file ckmessagelogging.C.

Referenced by _bufferedLocalMessageAckHandler(), and _messageLoggingInit().

void _bufferedLocalMessageAckHandler ( BufferedLocalLogHeader * recvdHeader )

Definition at line 1237 of file ckmessagelogging.C.

Referenced by _messageLoggingInit().

void _bufferedTicketRequestHandler ( BufferedTicketRequestHeader * recvdHeader )

Handler used for dealing with a bunch of ticket requests from one processor.

The replies are also bunched together Does not use _ticketRequestHandler

Definition at line 792 of file ckmessagelogging.C.

Referenced by _messageLoggingInit().

void _bufferedTicketHandler ( BufferedTicketRequestHeader * recvdHeader )

Message to handle the bunch of tickets that we get from one processor.

We send the tickets to be handled one at a time

Definition at line 1066 of file ckmessagelogging.C.

Referenced by _messageLoggingInit().

void _storeDeterminantsHandler ( char * buffer )

Stores the determinants coming from other processor.

Definition at line 861 of file ckcausalmlog.C.

References Converse::CkMyPe(), CmiFree(), CmiMemoryCheck(), Converse::CmiSyncSend(), RemoveDeterminantsHeader::index, StoreDeterminantsHeader::index, isSameDet(), n, StoreDeterminantsHeader::number, StoreDeterminantsHeader::PE, RemoveDeterminantsHeader::phase, StoreDeterminantsHeader::phase, CkVec< T >::push_back(), Determinant::receiver, Determinant::sender, CkVec< T >::size(), Determinant::SN, and Determinant::TN.

Referenced by _messageLoggingInit().

void _removeDeterminantsHandler ( char * buffer )

Removes the determinants after a particular index in the _localDets array.

Definition at line 834 of file ckcausalmlog.C.

References CmiFree(), RemoveDeterminantsHeader::index, and RemoveDeterminantsHeader::phase.

Referenced by _messageLoggingInit().

void _skipCldEnqueue	(	int	pe,
		envelope *	env,
		int	infoFn
	)

Definition at line 1361 of file ck.C.

void _noCldNodeEnqueue	(	int	node,
		envelope *	env
	)

Definition at line 1467 of file ck.C.

void generalCldEnqueue	(	int	destPE,
		envelope *	env,
		int	_infoIdx
	)

Definition at line 1143 of file ckcausalmlog.C.

void retryTicketRequest	(	void *	_ticketRequest,
		double	curWallTime
	)

Definition at line 1396 of file ckmessagelogging.C.

Referenced by _bufferedTicketRequestHandler(), _messageLoggingInit(), _ticketRequestHandler(), and retryTicketRequestTimer().

int preProcessReceivedMessage	(	envelope *	env,
		Chare **	objPointer,
		MlogEntry **	localLogEntry
	)

Definition at line 1012 of file ckcausalmlog.C.

void postProcessReceivedMessage	(	Chare *	obj,
		CkObjID &	sender,
		MCount	SN,
		MlogEntry *	entry
	)

Updates a few variables once a message has been processed.

Definition at line 1123 of file ckcausalmlog.C.

CpvExtern	(	StoredCheckpoint *	,
		_storedCheckpointData
	)

void checkpointAlarm	(	void *	_dummy,
		double	curWallTime
	)

Definition at line 1179 of file ckcausalmlog.C.

void startMlogCheckpoint	(	void *	_dummy,
		double	curWallTime
	)

Starts the checkpoint phase after migration.

Definition at line 1203 of file ckcausalmlog.C.

void pupArrayElementsSkip	(	PUP::er &	p,
		CmiBool	create,
		MigrationRecord *	listToSkip,
		int	listSize = `0`
	)

Pups all the array elements in this processor.

Definition at line 1313 of file ckcausalmlog.C.

void _checkpointRequestHandler ( CheckpointRequest * request )

Definition at line 1196 of file ckcausalmlog.C.

void _storeCheckpointHandler ( char * msg )

Definition at line 1386 of file ckcausalmlog.C.

void _checkpointAckHandler ( CheckPointAck * ackMsg )

Definition at line 1476 of file ckcausalmlog.C.

void _removeProcessedLogHandler ( char * requestMsg )

Removes messages in the log according to the received ticket numbers.

Definition at line 1588 of file ckcausalmlog.C.

void garbageCollectMlog ( )

Garbage collects the message log and other data structures.

In case of synchronized checkpoint, we use an optimization to avoid causal message logging protocol to communicate all determinants to the rest of the processors.

Definition at line 3308 of file ckcausalmlog.C.

References Converse::CkMyPe(), forAllCharesDo(), garbageCollectMlogForChare(), CkHashtableIterator::hasNext(), CkHashtableIterator::next(), and CkVec< T >::removeAll().

Referenced by initMlogLBStep().

void CkMlogRestart	(	const char *	dummy,
		CkArgMsg *	dummyMsg
	)

Function for restarting the crashed processor.

It sets the restart flag and contacts the buddy processor to get the latest checkpoint.

Definition at line 1652 of file ckcausalmlog.C.

void CkMlogRestartDouble	(	void *	,
		double
	)

Definition at line 1866 of file ckcausalmlog.C.

void processReceivedTN	(	Chare *	obj,
		int	vecsize,
		MCount *	listTNs
	)

Processes the received list of tickets from a particular PE.

Definition at line 2696 of file ckcausalmlog.C.

void processReceivedDet	(	Chare *	obj,
		int	vecsize,
		Determinant *	listDets
	)

Processes the received list of determinants from a particular PE.

Definition at line 2679 of file ckcausalmlog.C.

References Converse::CkMyPe(), CmiMemoryCheck(), Chare::mlogData, printDet(), Determinant::sender, Determinant::SN, Determinant::TN, and ChareMlogData::verifyTicket().

Referenced by _receivedDetDataHandler(), and _sendDetsReplyHandler().

void initializeRestart	(	void *	data,
		ChareMlogData *	mlogData
	)

Initializes variables and flags for restarting procedure.

Definition at line 2168 of file ckcausalmlog.C.

void distributeRestartedObjects ( )

Distributes objects to accelerate recovery after a failure.

Definition at line 2934 of file ckcausalmlog.C.

void sendDummyMigration	(	int	restartPE,
		CkGroupID	lbID,
		CkGroupID	locMgrID,
		CkArrayIndexMax &	idx,
		int	locationPE
	)

this method is used to send messages to a restarted processor to tell it that a particular expected object is not going to get to it

Definition at line 2980 of file ckcausalmlog.C.

void CkMlogRestartLocal ( )

Definition at line 1871 of file ckcausalmlog.C.

void _getCheckpointHandler ( RestartRequest * restartMsg )

Gets the stored checkpoint for its buddy processor.

Definition at line 1878 of file ckcausalmlog.C.

void _recvCheckpointHandler ( char * _restartData )

Receives the checkpoint data from its buddy, restores the state of all the objects and asks everyone else to update its home.

Definition at line 2036 of file ckcausalmlog.C.

void _resendMessagesHandler ( char * msg )

Resends messages since last checkpoint to the list of objects included in the request.

It also sends stored remote determinants to the particular failed PE.

Definition at line 2472 of file ckcausalmlog.C.

void _sendDetsHandler ( char * msg )

Send all remote determinants to a particular failed PE.

It only sends determinants to those objects on the list.

Definition at line 2369 of file ckcausalmlog.C.

References Converse::CkMyPe(), CmiAlloc(), CmiFree(), CmiMemoryCheck(), Converse::CmiSyncSendAndFree(), CmiWallTimer(), CkVec< T >::getVec(), ResendData::listObjects, ResendRequest::numberObjects, ResendData::numberObjects, ResendRequest::PE, ResendData::PE, printDet(), CkVec< T >::push_back(), TProcessedLog::recver, CkVec< T >::size(), ResendData::ticketVecs, and TProcessedLog::tProcessed.

Referenced by _messageLoggingInit().

void _sendDetsReplyHandler ( char * msg )

Receives determinants stored on remote nodes.

Definition at line 2528 of file ckcausalmlog.C.

References _resendMessagesHandler(), Converse::CkMyPe(), Converse::CkNumPes(), CmiAlloc(), CmiFree(), CmiMyPe(), Converse::CmiSyncSend(), Converse::CmiSyncSendAndFree(), CmiWallTimer(), createObjIDList(), distributeRestartedObjects(), fastRecovery, forAllCharesDo(), ResendRequest::numberObjects, ReceivedDetData::numDets, ReceivedTNData::numTNs, ResendRequest::PE, printf(), processReceivedDet(), processReceivedTN(), CentralLB::ReceiveDummyMigration(), ReceivedDetData::recver, and ReceivedTNData::recver.

Referenced by _messageLoggingInit(), and _recvGlobalStepHandler().

void _receivedTNDataHandler ( ReceivedTNData * msg )

Receives a list of TNs coming from the home PE of a migrated object (parallel restart).

Definition at line 2661 of file ckcausalmlog.C.

void _receivedDetDataHandler ( ReceivedDetData * msg )

Receives a list of determinants coming from the home PE of a migrated object (parallel restart).

Definition at line 2643 of file ckcausalmlog.C.

References CmiFree(), CmiMyPe(), Converse::CmiSyncSendAndFree(), CkObjID::getObject(), CkObjID::guessPE(), Chare::mlogData, ReceivedDetData::numDets, ChareMlogData::objID, printf(), processReceivedDet(), ReceivedDetData::recver, and CkObjID::toString().

Referenced by _messageLoggingInit().

void _distributedLocationHandler ( char * receivedMsg )

Handler to update information about an object just received.

Definition at line 2944 of file ckcausalmlog.C.

void _updateHomeRequestHandler ( RestartRequest * updateRequest )

Updates the homePe for all chares in this processor.

Definition at line 2201 of file ckcausalmlog.C.

void _updateHomeAckHandler ( RestartRequest * updateHomeAck )

Receives the updateHome ACKs from all other processors.

Once everybody has replied, it sends a request to resend the logged messages.

Definition at line 2114 of file ckcausalmlog.C.

void _verifyAckRequestHandler ( VerifyAckMsg * verifyRequest )

Definition at line 1915 of file ckcausalmlog.C.

void _verifyAckHandler ( VerifyAckMsg * verifyReply )

Definition at line 1943 of file ckcausalmlog.C.

void _dummyMigrationHandler ( DummyMigrationMsg * msg )

this handler is used to process a dummy migration msg.

it looks up the load balancer and calls migrated for it

Definition at line 3013 of file ckcausalmlog.C.

void _restartHandler ( RestartRequest * restartMsg )

Function to restart this processor.

The handler is invoked by a member of its same team in message logging.

Definition at line 1683 of file ckcausalmlog.C.

void _getRestartCheckpointHandler ( RestartRequest * restartMsg )

Gets the stored checkpoint but calls another function in the sender.

Definition at line 1715 of file ckcausalmlog.C.

void _recvRestartCheckpointHandler ( char * _restartData )

Receives the checkpoint coming from its buddy.

This is the case of restart for one team member that did not crash.

Definition at line 1754 of file ckcausalmlog.C.

void startLoadBalancingMlog	(	void()(void )	fnPtr,
		void *	_centralLb
	)

Load Balancing.

Definition at line 3105 of file ckcausalmlog.C.

void finishedCheckpointLoadBalancing ( )

Definition at line 3118 of file ckcausalmlog.C.

void sendMlogLocation	(	int	targetPE,
		envelope *	env
	)

Definition at line 3130 of file ckcausalmlog.C.

void resumeFromSyncRestart	(	void *	data,
		ChareMlogData *	mlogData
	)

Definition at line 3225 of file ckcausalmlog.C.

void _receiveMlogLocationHandler ( void * buf )

Definition at line 3211 of file ckcausalmlog.C.

void _receiveMigrationNoticeHandler ( MigrationNotice * msg )

Definition at line 3181 of file ckcausalmlog.C.

void _receiveMigrationNoticeAckHandler ( MigrationNoticeAck * msg )

Definition at line 3191 of file ckcausalmlog.C.

void _getGlobalStepHandler ( LBStepMsg * msg )

Definition at line 3394 of file ckcausalmlog.C.

void _recvGlobalStepHandler ( LBStepMsg * msg )

Receives the global step handler from PE 0.

Definition at line 3406 of file ckcausalmlog.C.

void _checkpointBarrierHandler ( CheckpointBarrierMsg * msg )

Processor 0 receives a contribution from every other processor after checkpoint.

Definition at line 3251 of file ckcausalmlog.C.

void _checkpointBarrierAckHandler ( CheckpointBarrierMsg * msg )

Definition at line 3270 of file ckcausalmlog.C.

int getCheckPointPE ( )

Getting the pe number of the current processor's buddy.

In the team-based approach each processor might checkpoint in the next team, but currently teams are only meant to reduce memory overhead.

Definition at line 3953 of file ckcausalmlog.C.

int isSameDet	(	Determinant *	first,
		Determinant *	second
	)			`[inline]`

Definition at line 3965 of file ckcausalmlog.C.

References Determinant::receiver, Determinant::sender, Determinant::SN, and Determinant::TN.

Referenced by _storeDeterminantsHandler().

void forAllCharesDo	(	MlogFn	fnPointer,
		void *	data
	)

Map function pointed by fnPointer over all the chares living in this processor.

Definition at line 3064 of file ckcausalmlog.C.

envelope* copyEnvelope ( envelope * env )

Definition at line 3958 of file ckcausalmlog.C.

void _resetNodeBocInitVec ( void )

void informLocationHome	(	CkGroupID	mgrID,
		CkArrayIndexMax	idx,
		int	homePE,
		int	currentPE
	)

method that informs an array elements home processor of its current location It is a converse method to bypass the charm++ message logging framework

Definition at line 3338 of file ckcausalmlog.C.

void _receiveLocationHandler ( CurrentLocationMsg * data )

Definition at line 3353 of file ckcausalmlog.C.

void CmiDeliverRemoteMsgHandlerRange	(	int	lowerHandler,
		int	higherHandler
	)

Referenced by processRemoteMlogMessages().

Variable Documentation

char objString[100]

Definition at line 159 of file ckcausalmlog.C.

int _maxBufferedTicketRequests

Definition at line 94 of file ckmessagelogging.C.

Referenced by _parseCommandLineOpts().

int _checkpointRequestHandlerIdx

Definition at line 160 of file ckcausalmlog.C.

int _storeCheckpointHandlerIdx

Definition at line 161 of file ckcausalmlog.C.

int _checkpointAckHandlerIdx

Definition at line 162 of file ckcausalmlog.C.

int _removeProcessedLogHandlerIdx

Definition at line 165 of file ckcausalmlog.C.

int _getCheckpointHandlerIdx

Definition at line 163 of file ckcausalmlog.C.

int _recvCheckpointHandlerIdx

Definition at line 164 of file ckcausalmlog.C.

int _resendMessagesHandlerIdx

Definition at line 177 of file ckcausalmlog.C.

int _sendDetsHandlerIdx

Definition at line 178 of file ckcausalmlog.C.

int _sendDetsReplyHandlerIdx

Definition at line 179 of file ckcausalmlog.C.

int _receivedTNDataHandlerIdx

Definition at line 180 of file ckcausalmlog.C.

int _receivedDetDataHandlerIdx

Definition at line 181 of file ckcausalmlog.C.

int _distributedLocationHandlerIdx

Definition at line 182 of file ckcausalmlog.C.

int _updateHomeRequestHandlerIdx

Definition at line 175 of file ckcausalmlog.C.

int _updateHomeAckHandlerIdx

Definition at line 176 of file ckcausalmlog.C.

int _verifyAckRequestHandlerIdx

Definition at line 167 of file ckcausalmlog.C.

int _verifyAckHandlerIdx

Definition at line 168 of file ckcausalmlog.C.

int _dummyMigrationHandlerIdx

Definition at line 169 of file ckcausalmlog.C.

int onGoingLoadBalancing

For testing on clusters we might carry out restarts on a porcessor without actually starting it 1 -> false restart 0 -> restart after an actual crash.

Definition at line 208 of file ckcausalmlog.C.

void* centralLb

Definition at line 209 of file ckcausalmlog.C.

void(* resumeLbFnPtr)(void *)

int _receiveMlogLocationHandlerIdx

Definition at line 211 of file ckcausalmlog.C.

int _receiveMigrationNoticeHandlerIdx

Definition at line 212 of file ckcausalmlog.C.

int _receiveMigrationNoticeAckHandlerIdx

Definition at line 213 of file ckcausalmlog.C.

int _getGlobalStepHandlerIdx

Definition at line 172 of file ckcausalmlog.C.

int _recvGlobalStepHandlerIdx

Definition at line 173 of file ckcausalmlog.C.

int _checkpointBarrierHandlerIdx

Definition at line 214 of file ckcausalmlog.C.

int _checkpointBarrierAckHandlerIdx

Definition at line 215 of file ckcausalmlog.C.

CkVec<MigrationRecord> migratedNoticeList

Definition at line 217 of file ckcausalmlog.C.

CkVec<RetainedMigratedObject *> retainedObjectList

Definition at line 218 of file ckcausalmlog.C.

int _receiveLocationHandlerIdx

Definition at line 232 of file ckcausalmlog.C.


Data Structures
struct	Determinant
	Struct to store the determinant of a particular message. More...
struct	RemoveDeterminantsHeader
	Struct for the header of the removeDeterminants handler. More...
struct	StoreDeterminantsHeader
	Struct for the header of the storeDeterminants handler. More...
class	Ticket
	Structure for a ticket assigned to a particular message. More...
class	SNToTicket
	Class that maps SN (sequence numbers) to TN (ticket numbers) for a particular object. More...
class	ChareMlogData
	This file includes the definition of the class for storing the meta data associdated with the message logging protocol. More...
class	MlogEntry
	Entry in a message log. More...
class	LocationID
	Class that represents the location of an array element. More...
class	StoredCheckpoint
class	RestoredLocalMap
	Class for storing metadata of local messages. More...
struct	TicketRequest
struct	TicketReply
struct	BufferedLocalLogHeader
struct	CheckPointDataMsg
struct	TProcessedLog
struct	RestartRequest
	Struct to request a particular action during restart. More...
struct	RestartProcessorData
struct	ResendRequest
struct	ReceivedTNData
struct	ReceivedDetData
struct	ResendData
struct	MigrationRecord
struct	MigrationNotice
struct	MigrationNoticeAck
struct	RetainedMigratedObject
struct	VerifyAckMsg
struct	CheckpointBarrierMsg
struct	CurrentLocationMsg
struct	LBStepMsg
struct	DummyMigrationMsg
Typedefs
typedef CkHashtableT < CkHashtableAdaptorT< CkObjID > , CkVec< Determinant > * >	CkDeterminantHashtableT
	Typedef for the hashtable type that maps object IDs to determinants.
typedef BufferedLocalLogHeader	BufferedTicketRequestHeader
typedef CheckPointDataMsg	CheckPointAck
typedef RestartRequest	CkPingMsg
typedef RestartRequest	CheckpointRequest
typedef ResendRequest	RemoveLogRequest
typedef void(*	MlogFn )(void , ChareMlogData )
Functions
	CpvExtern (Chare *, _currentObj)
	PUPbytes (Ticket) class MlogEntry
	CpvExtern (CkQ< TicketRequest * > *, _delayedTicketRequests)
	CpvExtern (CkQ< MlogEntry * > *, _delayedLocalTicketRequests)
	CpvExtern (char **, _bufferedTicketRequests)
void	_messageLoggingInit ()
	Initialize message logging data structures and register handlers.
void	sendGroupMsg (envelope *env, int destPE, int _infoIdx)
	Sends a group message that might be a broadcast.
void	sendArrayMsg (envelope *env, int destPE, int _infoIdx)
	Sends a message to an array element.
void	sendChareMsg (envelope env, int destPE, int _infoIdx, const CkChareID pCid)
	Sends a message to a singleton chare.
void	sendNodeGroupMsg (envelope *env, int destNode, int _infoIdx)
	Sends a nodegroup message that might be a broadcast.
void	sendCommonMsg (CkObjID &recver, envelope *env, int destPE, int _infoIdx)
	A method to generate the actual ticket requests for groups, nodegroups or arrays.
void	sendMsg (CkObjID &sender, CkObjID &recver, int destPE, MlogEntry *entry, MCount SN, MCount TN, int resend)
	Method that does the actual send by creating a ticket request filling it up and sending it.
void	sendLocalMsg (MlogEntry *entry)
	Function to send a local message.
void	_ticketRequestHandler (TicketRequest *)
	If there are any delayed requests, process them first before processing this request.
void	_ticketHandler (TicketReply *)
	This function handles the ticket received after a request.
void	_pingHandler (CkPingMsg *msg)
void	_bufferedLocalMessageCopyHandler (BufferedLocalLogHeader *recvdHeader, int freeHeader=1)
void	_bufferedLocalMessageAckHandler (BufferedLocalLogHeader *recvdHeader)
void	_bufferedTicketRequestHandler (BufferedTicketRequestHeader *recvdHeader)
	Handler used for dealing with a bunch of ticket requests from one processor.
void	_bufferedTicketHandler (BufferedTicketRequestHeader *recvdHeader)
	Message to handle the bunch of tickets that we get from one processor.
void	_storeDeterminantsHandler (char *buffer)
	Stores the determinants coming from other processor.
void	_removeDeterminantsHandler (char *buffer)
	Removes the determinants after a particular index in the _localDets array.
void	_skipCldEnqueue (int pe, envelope *env, int infoFn)
void	_noCldNodeEnqueue (int node, envelope *env)
void	generalCldEnqueue (int destPE, envelope *env, int _infoIdx)
void	retryTicketRequest (void *_ticketRequest, double curWallTime)
int	preProcessReceivedMessage (envelope env, Chare objPointer, MlogEntry *localLogEntry)
void	postProcessReceivedMessage (Chare obj, CkObjID &sender, MCount SN, MlogEntry entry)
	Updates a few variables once a message has been processed.
	CpvExtern (StoredCheckpoint *, _storedCheckpointData)
void	checkpointAlarm (void *_dummy, double curWallTime)
void	startMlogCheckpoint (void *_dummy, double curWallTime)
	Starts the checkpoint phase after migration.
void	pupArrayElementsSkip (PUP::er &p, CmiBool create, MigrationRecord *listToSkip, int listSize=0)
	Pups all the array elements in this processor.
void	_checkpointRequestHandler (CheckpointRequest *request)
void	_storeCheckpointHandler (char *msg)
void	_checkpointAckHandler (CheckPointAck *ackMsg)
void	_removeProcessedLogHandler (char *requestMsg)
	Removes messages in the log according to the received ticket numbers.
void	garbageCollectMlog ()
	Garbage collects the message log and other data structures.
void	CkMlogRestart (const char dummy, CkArgMsg dummyMsg)
	Function for restarting the crashed processor.
void	CkMlogRestartDouble (void *, double)
void	processReceivedTN (Chare obj, int vecsize, MCount listTNs)
	Processes the received list of tickets from a particular PE.
void	processReceivedDet (Chare obj, int vecsize, Determinant listDets)
	Processes the received list of determinants from a particular PE.
void	initializeRestart (void data, ChareMlogData mlogData)
	Initializes variables and flags for restarting procedure.
void	distributeRestartedObjects ()
	Distributes objects to accelerate recovery after a failure.
void	sendDummyMigration (int restartPE, CkGroupID lbID, CkGroupID locMgrID, CkArrayIndexMax &idx, int locationPE)
	this method is used to send messages to a restarted processor to tell it that a particular expected object is not going to get to it
void	CkMlogRestartLocal ()
void	_getCheckpointHandler (RestartRequest *restartMsg)
	Gets the stored checkpoint for its buddy processor.
void	_recvCheckpointHandler (char *_restartData)
	Receives the checkpoint data from its buddy, restores the state of all the objects and asks everyone else to update its home.
void	_resendMessagesHandler (char *msg)
	Resends messages since last checkpoint to the list of objects included in the request.
void	_sendDetsHandler (char *msg)
	Send all remote determinants to a particular failed PE.
void	_sendDetsReplyHandler (char *msg)
	Receives determinants stored on remote nodes.
void	_receivedTNDataHandler (ReceivedTNData *msg)
	Receives a list of TNs coming from the home PE of a migrated object (parallel restart).
void	_receivedDetDataHandler (ReceivedDetData *msg)
	Receives a list of determinants coming from the home PE of a migrated object (parallel restart).
void	_distributedLocationHandler (char *receivedMsg)
	Handler to update information about an object just received.
void	_updateHomeRequestHandler (RestartRequest *updateRequest)
	Updates the homePe for all chares in this processor.
void	_updateHomeAckHandler (RestartRequest *updateHomeAck)
	Receives the updateHome ACKs from all other processors.
void	_verifyAckRequestHandler (VerifyAckMsg *verifyRequest)
void	_verifyAckHandler (VerifyAckMsg *verifyReply)
void	_dummyMigrationHandler (DummyMigrationMsg *msg)
	this handler is used to process a dummy migration msg.
void	_restartHandler (RestartRequest *restartMsg)
	Function to restart this processor.
void	_getRestartCheckpointHandler (RestartRequest *restartMsg)
	Gets the stored checkpoint but calls another function in the sender.
void	_recvRestartCheckpointHandler (char *_restartData)
	Receives the checkpoint coming from its buddy.
void	startLoadBalancingMlog (void(fnPtr)(void ), void *_centralLb)
	Load Balancing.
void	finishedCheckpointLoadBalancing ()
void	sendMlogLocation (int targetPE, envelope *env)
void	resumeFromSyncRestart (void data, ChareMlogData mlogData)
void	_receiveMlogLocationHandler (void *buf)
void	_receiveMigrationNoticeHandler (MigrationNotice *msg)
void	_receiveMigrationNoticeAckHandler (MigrationNoticeAck *msg)
void	_getGlobalStepHandler (LBStepMsg *msg)
void	_recvGlobalStepHandler (LBStepMsg *msg)
	Receives the global step handler from PE 0.
void	_checkpointBarrierHandler (CheckpointBarrierMsg *msg)
	Processor 0 receives a contribution from every other processor after checkpoint.
void	_checkpointBarrierAckHandler (CheckpointBarrierMsg *msg)
int	getCheckPointPE ()
	Getting the pe number of the current processor's buddy.
int	isSameDet (Determinant first, Determinant second)
void	forAllCharesDo (MlogFn fnPointer, void *data)
	Map function pointed by fnPointer over all the chares living in this processor.
envelope *	copyEnvelope (envelope *env)
void	_initDone (void)
	This function (not a handler) is called once and only once per processor.
void	_resetNodeBocInitVec (void)
void	informLocationHome (CkGroupID mgrID, CkArrayIndexMax idx, int homePE, int currentPE)
	method that informs an array elements home processor of its current location It is a converse method to bypass the charm++ message logging framework
void	_receiveLocationHandler (CurrentLocationMsg *data)
void	CmiDeliverRemoteMsgHandlerRange (int lowerHandler, int higherHandler)
Variables
char	objString [100]
int	_maxBufferedTicketRequests
int	_checkpointRequestHandlerIdx
int	_storeCheckpointHandlerIdx
int	_checkpointAckHandlerIdx
int	_removeProcessedLogHandlerIdx
int	_getCheckpointHandlerIdx
int	_recvCheckpointHandlerIdx
int	_resendMessagesHandlerIdx
int	_sendDetsHandlerIdx
int	_sendDetsReplyHandlerIdx
int	_receivedTNDataHandlerIdx
int	_receivedDetDataHandlerIdx
int	_distributedLocationHandlerIdx
int	_updateHomeRequestHandlerIdx
int	_updateHomeAckHandlerIdx
int	_verifyAckRequestHandlerIdx
int	_verifyAckHandlerIdx
int	_dummyMigrationHandlerIdx
int	onGoingLoadBalancing
	For testing on clusters we might carry out restarts on a porcessor without actually starting it 1 -> false restart 0 -> restart after an actual crash.
void *	centralLb
void(*	resumeLbFnPtr )(void *)
int	_receiveMlogLocationHandlerIdx
int	_receiveMigrationNoticeHandlerIdx
int	_receiveMigrationNoticeAckHandlerIdx
int	_getGlobalStepHandlerIdx
int	_recvGlobalStepHandlerIdx
int	_checkpointBarrierHandlerIdx
int	_checkpointBarrierAckHandlerIdx
CkVec< MigrationRecord >	migratedNoticeList
CkVec< RetainedMigratedObject * >	retainedObjectList
int	_receiveLocationHandlerIdx