ck-core/ckcausalmlog.C File Reference

Go to the source code of this file.

Data Structures

class ElementPacker

class ElementDistributor

class ElementCaller

Functions

const char * idx2str (const CkArrayIndex &ind)

const char * idx2str (const ArrayElement *el)

void getGlobalStep (CkGroupID gID)

bool fault_aware (CkObjID &recver)

void sendCheckpointData (int mode)

Sends the checkpoint to its buddy.

void createObjIDList (void *data, ChareMlogData *mlogData)

bool isLocal (int destPE)

Determines if the message is local or not.

bool isTeamLocal (int destPE)

Determines if the message is group local or not.

void printLog (TProcessedLog *log)

Prints a processed log.

void readKillFile ()

CpvDeclare (Chare *, _currentObj)

CpvDeclare (StoredCheckpoint *, _storedCheckpointData)

CpvDeclare (CkQ< MlogEntry * > *, _delayedLocalMsgs)

CpvDeclare (Queue, _outOfOrderMessageQueue)

CpvDeclare (Queue, _delayedRemoteMessageQueue)

CpvDeclare (char **, _bufferedTicketRequests)

CpvDeclare (int *, _numBufferedTicketRequests)

CpvDeclare (char *, _localDets)

CpvDeclare (CkDeterminantHashtableT *, _remoteDets)

CpvDeclare (char *, _incarnation)

CpvDeclare (CkVec< LocationID * > *, _emigrantRecObjs)

CpvDeclare (CkVec< CkLocRec_local * > *, _immigrantRecObjs)

void setTeamRecovery (void *data, ChareMlogData *mlogData)

Turns on the flag for team recovery that selectively restores particular metadata information.

void unsetTeamRecovery (void *data, ChareMlogData *mlogData)

Turns off the flag for team recovery.

void _messageLoggingInit ()

Initialize message logging data structures and register handlers.

void killLocal (void *_dummy, double curWallTime)

void readFaultFile ()

: reads the PE that will be failing throughout the execution and the mean time between failures.

void addBufferedDeterminant (CkObjID sender, CkObjID receiver, MCount SN, MCount TN)

Adds a determinants to the buffered determinants and checks whether the array of buffered determinants needs to be extended.

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

Sends a group message that might be a broadcast.

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

Sends a nodegroup 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 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 _removeDeterminantsHandler (char *buffer)

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

void _storeDeterminantsHandler (char *buffer)

Stores the determinants coming from other processor.

void _ticketRequestHandler (TicketRequest *ticketRequest)

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

bool _getTicket (envelope *env, int *flag)

Gets a ticket for a recently received message.

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

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

Updates a few variables once a message has been processed.

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

void _pingHandler (CkPingMsg *msg)

void buildProcessedTicketLog (void *data, ChareMlogData *mlogData)

A chare adds the latest ticket number processed.

void clearUpMigratedRetainedLists (int PE)

void checkpointAlarm (void *_dummy, double curWallTime)

void _checkpointRequestHandler (CheckpointRequest *request)

void startMlogCheckpoint (void *_dummy, double curWallTime)

Starts the checkpoint phase after migration.

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

Pups all the array elements in this processor.

void writeCheckpointToDisk (int size, char *chkpt)

void _storeCheckpointHandler (char *msg)

void sendRemoveLogRequests ()

Sends out the messages asking senders to throw away message logs below a certain ticket number.

void _checkpointAckHandler (CheckPointAck *ackMsg)

void populateDeterminantTable (char *data)

Inserts all the determinants into a hash table.

void removeProcessedLogs (void *_data, ChareMlogData *mlogData)

void _removeProcessedLogHandler (char *requestMsg)

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

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

Function for restarting the crashed processor.

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 CkMlogRestartDouble (void *, double)

void CkMlogRestartLocal ()

void _getCheckpointHandler (RestartRequest *restartMsg)

Gets the stored checkpoint for its buddy processor.

void _verifyAckRequestHandler (VerifyAckMsg *verifyRequest)

void _verifyAckHandler (VerifyAckMsg *verifyReply)

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 _updateHomeAckHandler (RestartRequest *updateHomeAck)

Receives the updateHome ACKs from all other processors.

void initializeRestart (void *data, ChareMlogData *mlogData)

Initializes variables and flags for restarting procedure.

void updateHomePE (void *data, ChareMlogData *mlogData)

Updates the homePe of chare array elements.

void _updateHomeRequestHandler (RestartRequest *updateRequest)

Updates the homePe for all chares in this processor.

void fillTicketForChare (void *data, ChareMlogData *mlogData)

Fills up the ticket vector for each chare.

void printMsg (envelope *env, const char *par)

Prints information about a message.

void printDet (Determinant *det, const char *par)

Prints information about a determinant.

void resendMessageForChare (void *data, ChareMlogData *mlogData)

Resends all the logged messages to a particular chare list.

void _sendDetsHandler (char *msg)

Send all remote determinants to a particular failed PE.

void _resendMessagesHandler (char *msg)

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

MCount maxVec (CkVec< MCount > *TNvec)

Returns the maximum ticket from a vector.

void sortVec (CkVec< MCount > *TNvec)

int searchVec (CkVec< MCount > *TNVec, MCount searchTN)

void processDelayedRemoteMsgQueue ()

Processes the messages in the delayed remote message queue.

void _sendDetsReplyHandler (char *msg)

Receives determinants stored on remote nodes.

void _receivedDetDataHandler (ReceivedDetData *msg)

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

void _receivedTNDataHandler (ReceivedTNData *msg)

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

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

Processes the received list of determinants from a particular PE.

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

Processes the received list of tickets from a particular PE.

void distributeRestartedObjects ()

Distributes objects to accelerate recovery after a failure.

void _distributedLocationHandler (char *receivedMsg)

Handler to update information about an object just received.

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 sendDummyMigrationCounts (int *dummyCounts)

this method is used by a restarted processor to tell other processors that they are not going to receive these many objects.

void _dummyMigrationHandler (DummyMigrationMsg *msg)

this handler is used to process a dummy migration msg.

void forAllCharesDo (MlogFn fnPointer, void *data)

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

void initMlogLBStep (CkGroupID gid)

This is the first time Converse is called after AtSync method has been called by every local object.

void startLoadBalancingMlog (void(*_fnPtr)(void *), void *_centralLb)

Load Balancing.

void finishedCheckpointLoadBalancing ()

void sendMlogLocation (int targetPE, envelope *env)

void _receiveMigrationNoticeHandler (MigrationNotice *msg)

void _receiveMigrationNoticeAckHandler (MigrationNoticeAck *msg)

void _receiveMlogLocationHandler (void *buf)

void resumeFromSyncRestart (void *data, ChareMlogData *mlogData)

void checkAndSendCheckpointBarrierAcks (CheckpointBarrierMsg *msg)

Processor 0 sends a broadcast to every other processor after checkpoint barrier.

void _checkpointBarrierHandler (CheckpointBarrierMsg *msg)

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

void _checkpointBarrierAckHandler (CheckpointBarrierMsg *msg)

void garbageCollectMlogForChare (void *data, ChareMlogData *mlogData)

Function to remove all messages in the message log of a particular chare.

void garbageCollectMlog ()

Garbage collects the message log and other data structures.

void informLocationHome (CkGroupID locMgrID, 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 _getGlobalStepHandler (LBStepMsg *msg)

void _recvGlobalStepHandler (LBStepMsg *msg)

Receives the global step handler from PE 0.

void _messageLoggingExit ()

Function to wrap up performance information.

int getCheckPointPE ()

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

envelope * copyEnvelope (envelope *env)

int isSameDet (Determinant *first, Determinant *second)

Variables

int _restartFlag = 0

int _numRestartResponses = 0

int countHashRefs = 0

int countHashCollisions = 0

char * checkpointDirectory = "."

int unAckedCheckpoint = 0

int countLocal = 0

int countBuffered = 0

int countPiggy = 0

int countClearBufferedLocalCalls = 0

int countUpdateHomeAcks = 0

int teamSize

int chkptPeriod

bool fastRecovery

int parallelRecovery

char * killFile

char * faultFile

int killFlag = 0

int faultFlag = 0

int restartingMlogFlag = 0

double killTime = 0.0

double faultMean

int checkpointCount = 0

int _numBufferedDets

int _indexBufferedDets

int _phaseBufferedDets

int _maxBufferedDets

int * numMsgsTarget

int * sizeMsgsTarget

int totalMsgsTarget

float totalMsgsSize

int numPiggyDets

int numDets

int numDupDets

int msgLogSize

int bufferedDetsSize

int storedDetsSize

float MLOGFT_totalLogSize = 0.0

float MLOGFT_totalMessages = 0.0

static double adjustChkptPeriod = 0.0

static double nextCheckpointTime = 0.0

static CkHashtableT
< CkHashtableAdaptorT< CkObjID >
, CkHashtableT
< CkHashtableAdaptorT< CkObjID >
, SNToTicket * > * > detTable (1000, 0.3)

int _pingHandlerIdx

char objString [100]

int _checkpointRequestHandlerIdx

int _storeCheckpointHandlerIdx

int _checkpointAckHandlerIdx

int _getCheckpointHandlerIdx

int _recvCheckpointHandlerIdx

int _removeProcessedLogHandlerIdx

int _verifyAckRequestHandlerIdx

int _verifyAckHandlerIdx

int _dummyMigrationHandlerIdx

int _getGlobalStepHandlerIdx

int _recvGlobalStepHandlerIdx

int _updateHomeRequestHandlerIdx

int _updateHomeAckHandlerIdx

int _resendMessagesHandlerIdx

int _sendDetsHandlerIdx

int _sendDetsReplyHandlerIdx

int _receivedTNDataHandlerIdx

int _receivedDetDataHandlerIdx

int _distributedLocationHandlerIdx

int _storeDeterminantsHandlerIdx

int _removeDeterminantsHandlerIdx

int _restartHandlerIdx

int _getRestartCheckpointHandlerIdx

int _recvRestartCheckpointHandlerIdx

int verifyAckTotal

int verifyAckCount

int verifyAckedRequests = 0

RestartRequest * storedRequest

int _falseRestart = 0

int onGoingLoadBalancing = 0

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 _checkpointBarrierHandlerIdx

int _checkpointBarrierAckHandlerIdx

CkVec< MigrationRecord > migratedNoticeList

CkVec< RetainedMigratedObject * > retainedObjectList

int donotCountMigration = 0

int countLBMigratedAway = 0

int countLBToMigrate = 0

int migrationDoneCalled = 0

int checkpointBarrierCount = 0

int globalResumeCount = 0

CkGroupID globalLBID

int restartDecisionNumber = -1

double lastCompletedAlarm = 0

double lastRestart = 0

int _receiveLocationHandlerIdx

int calledRetryTicketRequest = 0

This method is used to retry the ticket requests that had been queued up earlier.

CkVec< TProcessedLog > processedTicketLog

double totalSearchRestoredTime = 0

double totalSearchRestoredCount = 0

Function Documentation

const char * idx2str ( const CkArrayIndex & ind )

Definition at line 28 of file ckmessagelogging.C.

References idx2str().

const char* idx2str ( const ArrayElement * el )

void getGlobalStep ( CkGroupID gID )

Definition at line 3384 of file ckcausalmlog.C.

References CmiMyPe(), Converse::CmiSyncSend(), LBStepMsg::fromPE, LBStepMsg::lbID, and LBStepMsg::step.

Referenced by _recvCheckpointHandler().

bool fault_aware ( CkObjID & recver )

Definition at line 996 of file ckcausalmlog.C.

References CkObjID::type, TypeArray, TypeChare, TypeGroup, TypeMainChare, and TypeNodeGroup.

Referenced by _removeProcessedLogHandler(), and preProcessReceivedMessage().

void sendCheckpointData ( int mode )

Sends the checkpoint to its buddy.

The mode distinguishes between the two cases: MLOG_RESTARTED: sending the checkpoint to a team member that did not crash but is restarting. MLOG_CRASHED: sending the checkpoint to the processor that crashed.

Definition at line 1959 of file ckcausalmlog.C.

References StoredCheckpoint::buf, StoredCheckpoint::bufSize, RestartProcessorData::checkPointSize, Converse::CkMyPe(), CmiAlloc(), CmiFree(), CmiMyPe(), Converse::CmiSyncSendAndFree(), CmiTimer(), CkVec< T >::getVec(), RestartProcessorData::lbGroupID, RestartProcessorData::migratedElementSize, msg, RestartProcessorData::numMigratedAwayElements, RestartProcessorData::numMigratedInElements, RestartProcessorData::PE, RestartRequest::PE, printf(), RestartProcessorData::restartWallTime, and CkVec< T >::size().

Referenced by _getCheckpointHandler(), _getRestartCheckpointHandler(), and _verifyAckHandler().

void createObjIDList	(	void *	data,
		ChareMlogData *	mlogData
	)

Definition at line 2020 of file ckcausalmlog.C.

References Converse::CkMyPe(), list, ChareMlogData::objID, objString, printLog(), CkVec< T >::push_back(), TProcessedLog::recver, CkObjID::toString(), ChareMlogData::tProcessed, and TProcessedLog::tProcessed.

Referenced by _recvCheckpointHandler(), _recvRestartCheckpointHandler(), _sendDetsReplyHandler(), and _updateHomeAckHandler().

bool isLocal ( int destPE ) [inline]

Determines if the message is local or not.

A message is local if: 1) Both the destination and origin are the same PE.

Definition at line 660 of file ckcausalmlog.C.

References Converse::CkMyPe().

Referenced by generateCommonTicketRequest(), and sendCommonMsg().

bool isTeamLocal ( int destPE ) [inline]

Determines if the message is group local or not.

A message is group local if: 1) They belong to the same group in the group-based message logging.

Definition at line 672 of file ckcausalmlog.C.

References Converse::CkMyPe().

Referenced by _resendMessagesHandler(), _ticketHandler(), generateCommonTicketRequest(), sendCommonMsg(), sendMsg(), and sendTicketRequest().

void printLog ( TProcessedLog * log )

Prints a processed log.

Definition at line 2287 of file ckcausalmlog.C.

References Converse::CkMyPe(), TProcessedLog::recver, CkObjID::toString(), and TProcessedLog::tProcessed.

Referenced by createObjIDList(), and KillOnAllSigs().

void readKillFile ( )

Definition at line 380 of file ckcausalmlog.C.

References CcdCallFnAfter(), Converse::CkMyPe(), CmiWallTimer(), killFile, killLocal(), and printf().

Referenced by _initCharm(), and CkMemCheckPT::isMaster().

CpvDeclare	(	Chare *	,
		_currentObj
	)

CpvDeclare	(	StoredCheckpoint *	,
		_storedCheckpointData
	)

CpvDeclare	(	CkQ< MlogEntry * > *	,
		_delayedLocalMsgs
	)

CpvDeclare	(	Queue	,
		_outOfOrderMessageQueue
	)

CpvDeclare	(	Queue	,
		_delayedRemoteMessageQueue
	)

CpvDeclare	(	char **	,
		_bufferedTicketRequests
	)

CpvDeclare	(	int *	,
		_numBufferedTicketRequests
	)

CpvDeclare	(	char *	,
		_localDets
	)

Note:: All the determinants generated by a PE are stored in variable _localDets. As soon as a message is sent, then all the determinants are appended to the message, but those determinants are not deleted. We must wait until an ACK comes from the receiver to delete the determinants. In the meantime the same determinants may be appended to other messages and more determinants can be added to _localDets. A simple solution to this problem was to have a primitive array and keep adding determinants at the end. However, to avoid multiple copies of determinants, we will keep a pointer to the first 'valid' determinant in the array. Alternatively, we can keep a pointer to the latest determinant and a number of how many valid determinants there are behind it. We do not remove determinants until a checkpoint is made, since these determinants may have to be added to messages in case of a recovery.

CpvDeclare	(	CkDeterminantHashtableT *	,
		_remoteDets
	)

CpvDeclare	(	char *	,
		_incarnation
	)

CpvDeclare	(	CkVec< LocationID * > *	,
		_emigrantRecObjs
	)

CpvDeclare	(	CkVec< CkLocRec_local * > *	,
		_immigrantRecObjs
	)

void setTeamRecovery	(	void *	data,
		ChareMlogData *	mlogData
	)

Turns on the flag for team recovery that selectively restores particular metadata information.

Definition at line 2272 of file ckcausalmlog.C.

References ChareMlogData::teamRecoveryFlag.

Referenced by _recvRestartCheckpointHandler().

void unsetTeamRecovery	(	void *	data,
		ChareMlogData *	mlogData
	)

Turns off the flag for team recovery.

Definition at line 2280 of file ckcausalmlog.C.

References ChareMlogData::teamRecoveryFlag.

Referenced by _recvRestartCheckpointHandler().

void _messageLoggingInit ( )

Initialize message logging data structures and register handlers.

Definition at line 237 of file ckcausalmlog.C.

Referenced by _initCharm().

void killLocal	(	void *	_dummy,
		double	curWallTime
	)

Definition at line 417 of file ckcausalmlog.C.

References CcdCallFnAfter(), Converse::CkMyPe(), CmiWallTimer(), and printf().

Referenced by CkMemCheckPT::isMaster(), readFaultFile(), and readKillFile().

void readFaultFile ( )

: reads the PE that will be failing throughout the execution and the mean time between failures.

We assume an exponential distribution for the mean-time-between-failures.

Definition at line 401 of file ckcausalmlog.C.

References CcdCallFnAfter(), Converse::CkMyPe(), faultFile, killLocal(), and printf().

void addBufferedDeterminant	(	CkObjID	sender,
		CkObjID	receiver,
		MCount	SN,
		MCount	TN
	)			`[inline]`

Adds a determinants to the buffered determinants and checks whether the array of buffered determinants needs to be extended.

Definition at line 432 of file ckcausalmlog.C.

References Converse::CkMyPe(), CmiAlloc(), CmiFree(), Determinant::receiver, Determinant::sender, Determinant::SN, and Determinant::TN.

Referenced by preProcessReceivedMessage().

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 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 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 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 _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 _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 _ticketRequestHandler ( TicketRequest * ticketRequest ) [inline]

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

Definition at line 928 of file ckcausalmlog.C.

Referenced by _messageLoggingInit().

bool _getTicket	(	envelope *	env,
		int *	flag
	)			`[inline]`

Gets a ticket for a recently received message.

Precondition:: env->recver has to be on this processor.

Returns:: Returns true if ticket assignment is successful, otherwise returns false. A false result is due to the fact that we are recovering.

Definition at line 941 of file ckcausalmlog.C.

References Converse::CkMyPe(), CmiMemoryCheck(), CmiWallTimer(), CkObjID::getObject(), ChareMlogData::getTicket(), Chare::mlogData, ChareMlogData::next_ticket(), printf(), envelope::recver, ChareMlogData::restartFlag, envelope::sender, envelope::SN, Ticket::state, Ticket::TN, envelope::TN, CkObjID::toString(), ChareMlogData::tProcessed, and ChareMlogData::verifyTicket().

Referenced by preProcessReceivedMessage().

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

Definition at line 1012 of file ckcausalmlog.C.

Referenced by _processHandler().

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.

Referenced by _processHandler().

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

Definition at line 1143 of file ckcausalmlog.C.

Referenced by _ticketHandler(), generateCommonTicketRequest(), sendCommonMsg(), and sendMsg().

void _pingHandler ( CkPingMsg * msg )

Definition at line 1164 of file ckcausalmlog.C.

Referenced by _messageLoggingInit().

void buildProcessedTicketLog	(	void *	data,
		ChareMlogData *	mlogData
	)

A chare adds the latest ticket number processed.

Definition at line 1283 of file ckcausalmlog.C.

References Converse::CkMyPe(), ChareMlogData::objID, objString, printf(), CkVec< T >::push_back(), TProcessedLog::recver, CkObjID::toString(), ChareMlogData::tProcessed, and TProcessedLog::tProcessed.

Referenced by startMlogCheckpoint().

void clearUpMigratedRetainedLists ( int PE )

Definition at line 1614 of file ckcausalmlog.C.

References CmiFree(), CmiMemoryCheck(), CmiMyPe(), RetainedMigratedObject::msg, printf(), CkVec< T >::remove(), and CkVec< T >::size().

Referenced by _removeProcessedLogHandler(), and sendRemoveLogRequests().

void checkpointAlarm	(	void *	_dummy,
		double	curWallTime
	)

Definition at line 1179 of file ckcausalmlog.C.

Referenced by checkpointAlarm(), and startMlogCheckpoint().

void _checkpointRequestHandler ( CheckpointRequest * request )

Definition at line 1196 of file ckcausalmlog.C.

Referenced by _messageLoggingInit().

void startMlogCheckpoint	(	void *	_dummy,
		double	curWallTime
	)

Starts the checkpoint phase after migration.

Definition at line 1203 of file ckcausalmlog.C.

Referenced by _checkpointRequestHandler(), _receiveMigrationNoticeAckHandler(), _updateHomeRequestHandler(), and startLoadBalancingMlog().

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

Pups all the array elements in this processor.

Definition at line 1313 of file ckcausalmlog.C.

Referenced by _recvCheckpointHandler(), _recvRestartCheckpointHandler(), and startMlogCheckpoint().

void writeCheckpointToDisk	(	int	size,
		char *	chkpt
	)

Definition at line 1369 of file ckcausalmlog.C.

References checkpointDirectory, Converse::CkMyPe(), and write().

Referenced by _storeCheckpointHandler().

void _storeCheckpointHandler ( char * msg )

Definition at line 1386 of file ckcausalmlog.C.

Referenced by _messageLoggingInit().

void sendRemoveLogRequests ( )

Sends out the messages asking senders to throw away message logs below a certain ticket number.

Note:: The remove log request message looks like |RemoveLogRequest||List of TProcessedLog||Number of Determinants||List of Determinants|

Definition at line 1437 of file ckcausalmlog.C.

References _startTime, Converse::CkMyPe(), Converse::CkNumPes(), clearUpMigratedRetainedLists(), CmiAbort(), CmiAlloc(), CmiFree(), CmiMemoryCheck(), CmiMyPe(), Converse::CmiSyncSend(), CkVec< T >::getVec(), ResendRequest::numberObjects, ResendRequest::PE, CkVec< T >::size(), and traceUserBracketEvent().

Referenced by _checkpointAckHandler(), and _checkpointBarrierAckHandler().

void _checkpointAckHandler ( CheckPointAck * ackMsg )

Definition at line 1476 of file ckcausalmlog.C.

Referenced by _messageLoggingInit().

void populateDeterminantTable ( char * data ) [inline]

Inserts all the determinants into a hash table.

Definition at line 1495 of file ckcausalmlog.C.

References CmiMemoryCheck(), detTable, CkHashtableT< KEY, OBJ >::get(), list, ResendRequest::numberObjects, SNToTicket::put(), CkHashtableTslow< KEY, OBJ >::put(), Determinant::TN, and Ticket::TN.

Referenced by _removeProcessedLogHandler().

void removeProcessedLogs	(	void *	_data,
		ChareMlogData *	mlogData
	)

Definition at line 1531 of file ckcausalmlog.C.

References Converse::CkMyPe(), CmiMemoryCheck(), data, CkQ< T >::deq(), detTable, CkQ< T >::enq(), MlogEntry::env, SNToTicket::get(), CkHashtableT< KEY, OBJ >::get(), ChareMlogData::getMlog(), CkQ< T >::length(), list, match(), ResendRequest::numberObjects, ChareMlogData::objID, printf(), TProcessedLog::recver, envelope::recver, envelope::sender, envelope::SN, envelope::TN, Ticket::TN, CkObjID::toString(), and TProcessedLog::tProcessed.

Referenced by _removeProcessedLogHandler().

void _removeProcessedLogHandler ( char * requestMsg )

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

Definition at line 1588 of file ckcausalmlog.C.

Referenced by _messageLoggingInit().

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.

Referenced by _initCharm(), _parseCommandLineOpts(), CkMlogRestartDouble(), and CkMlogRestartLocal().

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.

Referenced by _messageLoggingInit().

void _getRestartCheckpointHandler ( RestartRequest * restartMsg )

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

Definition at line 1715 of file ckcausalmlog.C.

Referenced by _messageLoggingInit().

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.

Referenced by _messageLoggingInit().

void CkMlogRestartDouble	(	void *	,
		double
	)

Definition at line 1866 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.

Referenced by _messageLoggingInit().

void _verifyAckRequestHandler ( VerifyAckMsg * verifyRequest )

Definition at line 1915 of file ckcausalmlog.C.

Referenced by _messageLoggingInit().

void _verifyAckHandler ( VerifyAckMsg * verifyReply )

Definition at line 1943 of file ckcausalmlog.C.

Referenced by _messageLoggingInit().

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.

Referenced by _messageLoggingInit().

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.

Referenced by _messageLoggingInit(), and _recvGlobalStepHandler().

void initializeRestart	(	void *	data,
		ChareMlogData *	mlogData
	)

Initializes variables and flags for restarting procedure.

Definition at line 2168 of file ckcausalmlog.C.

Referenced by _recvCheckpointHandler(), and _recvRestartCheckpointHandler().

void updateHomePE	(	void *	data,
		ChareMlogData *	mlogData
	)

Updates the homePe of chare array elements.

Definition at line 2177 of file ckcausalmlog.C.

References _ObjectID::array, CkArrayIndexBase::asChild(), CkArrayID::ckLocalBranch(), Converse::CkMyPe(), CkObjID::data, CkLocMgr::getGroupID(), CkArray::getLocMgr(), CkLocMgr::homePe(), _ObjectID::s_array::id, _ObjectID::s_array::idx, informLocationHome(), ChareMlogData::objID, RestartRequest::PE, CkArrayIndex::print(), printf(), CkObjID::type, and TypeArray.

Referenced by _updateHomeRequestHandler().

void _updateHomeRequestHandler ( RestartRequest * updateRequest )

Updates the homePe for all chares in this processor.

Definition at line 2201 of file ckcausalmlog.C.

Referenced by _messageLoggingInit().

void fillTicketForChare	(	void *	data,
		ChareMlogData *	mlogData
	)

Fills up the ticket vector for each chare.

Definition at line 2230 of file ckcausalmlog.C.

References SNToTicket::get(), SNToTicket::getFinishSN(), SNToTicket::getStartSN(), CkHashtable::iterator(), ResendData::listObjects, CkHashtableIterator::next(), ResendData::numberObjects, ResendData::PE, CkVec< T >::push_back(), ChareMlogData::teamTable, ResendData::ticketVecs, and Ticket::TN.

Referenced by _resendMessagesHandler().

void printMsg	(	envelope *	env,
		const char *	par
	)

Prints information about a message.

Definition at line 2295 of file ckcausalmlog.C.

References Converse::CkMyPe(), envelope::recver, envelope::sender, envelope::SN, and CkObjID::toString().

Referenced by processDelayedRemoteMsgQueue(), and resendMessageForChare().

void printDet	(	Determinant *	det,
		const char *	par
	)

Prints information about a determinant.

Definition at line 2304 of file ckcausalmlog.C.

References Converse::CkMyPe(), Determinant::receiver, Determinant::sender, Determinant::SN, Determinant::TN, and CkObjID::toString().

Referenced by _sendDetsHandler(), and processReceivedDet().

void resendMessageForChare	(	void *	data,
		ChareMlogData *	mlogData
	)

Resends all the logged messages to a particular chare list.

Parameters:

	data	is of type ResendData which contains the array of objects on the restartedProcessor.
	mlogData	a particular chare living in this processor.

Definition at line 2315 of file ckcausalmlog.C.

References Converse::CkMyPe(), Converse::CmiSyncSend(), copyEnvelope(), CqsEnqueueGeneral(), MlogEntry::env, ChareMlogData::getMlog(), envelope::getPriobits(), envelope::getPrioPtr(), envelope::getQueueing(), envelope::getTotalsize(), CkQ< T >::length(), ResendData::listObjects, ResendData::numberObjects, ChareMlogData::objID, ResendData::PE, printf(), printMsg(), envelope::recver, envelope::sender, envelope::SN, envelope::TN, CkObjID::toString(), CkObjID::type, TypeInvalid, and TypeNodeGroup.

Referenced by _resendMessagesHandler().

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 _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.

Referenced by _messageLoggingInit(), _recvRestartCheckpointHandler(), _sendDetsReplyHandler(), and _updateHomeAckHandler().

MCount maxVec ( CkVec< MCount > * TNvec )

Returns the maximum ticket from a vector.

Definition at line 2784 of file ckcausalmlog.C.

References max(), and CkVec< T >::size().

Referenced by processReceivedTN().

void sortVec ( CkVec< MCount > * TNvec )

Definition at line 2793 of file ckcausalmlog.C.

References CkVec< T >::push_back(), CkVec< T >::removeAll(), and CkVec< T >::size().

Referenced by processReceivedTN().

int searchVec	(	CkVec< MCount > *	TNVec,
		MCount	searchTN
	)

Definition at line 2824 of file ckcausalmlog.C.

References left(), size, and CkVec< T >::size().

Referenced by processReceivedTN().

void processDelayedRemoteMsgQueue ( )

Processes the messages in the delayed remote message queue.

Definition at line 2509 of file ckcausalmlog.C.

References Converse::CkMyPe(), CmiMemoryCheck(), CqsDequeue(), CqsEmpty(), CqsEnqueueGeneral(), envelope::getPriobits(), envelope::getPrioPtr(), printf(), and printMsg().

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 _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 _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.

Referenced by _messageLoggingInit().

void processReceivedDet	(	Chare *	obj,
		int	listSize,
		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 processReceivedTN	(	Chare *	obj,
		int	listSize,
		MCount *	listTNs
	)

Processes the received list of tickets from a particular PE.

Definition at line 2696 of file ckcausalmlog.C.

Referenced by _receivedTNDataHandler(), _resendReplyHandler(), and _sendDetsReplyHandler().

void distributeRestartedObjects ( )

Distributes objects to accelerate recovery after a failure.

Definition at line 2934 of file ckcausalmlog.C.

Referenced by _sendDetsReplyHandler(), and _updateHomeAckHandler().

void _distributedLocationHandler ( char * receivedMsg )

Handler to update information about an object just received.

Definition at line 2944 of file ckcausalmlog.C.

Referenced by _messageLoggingInit().

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.

Referenced by _resendMessagesHandler().

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.

Referenced by _messageLoggingInit().

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.

Referenced by _recvCheckpointHandler(), _recvRestartCheckpointHandler(), _removeProcessedLogHandler(), _resendMessagesHandler(), _sendDetsReplyHandler(), _updateHomeAckHandler(), _updateHomeRequestHandler(), garbageCollectMlog(), and startMlogCheckpoint().

void startLoadBalancingMlog	(	void()(void )	_fnPtr,
		void *	_centralLb
	)

Load Balancing.

Definition at line 3105 of file ckcausalmlog.C.

Referenced by CentralLB::MigrationDone().

void finishedCheckpointLoadBalancing ( )

Definition at line 3118 of file ckcausalmlog.C.

Referenced by _checkpointAckHandler().

void sendMlogLocation	(	int	targetPE,
		envelope *	env
	)

Definition at line 3130 of file ckcausalmlog.C.

Referenced by CkLocMgr::emigrate().

void _receiveMigrationNoticeHandler ( MigrationNotice * msg )

Definition at line 3181 of file ckcausalmlog.C.

Referenced by _messageLoggingInit().

void _receiveMigrationNoticeAckHandler ( MigrationNoticeAck * msg )

Definition at line 3191 of file ckcausalmlog.C.

Referenced by _messageLoggingInit().

void _receiveMlogLocationHandler ( void * buf )

Definition at line 3211 of file ckcausalmlog.C.

Referenced by _messageLoggingInit(), and _resendMessagesHandler().

void resumeFromSyncRestart	(	void *	data,
		ChareMlogData *	mlogData
	)

Definition at line 3225 of file ckcausalmlog.C.

void checkAndSendCheckpointBarrierAcks ( CheckpointBarrierMsg * msg ) [inline]

Processor 0 sends a broadcast to every other processor after checkpoint barrier.

Definition at line 3239 of file ckcausalmlog.C.

References Converse::CmiSyncSend().

Referenced by _checkpointBarrierHandler().

void _checkpointBarrierHandler ( CheckpointBarrierMsg * msg )

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

Definition at line 3251 of file ckcausalmlog.C.

Referenced by _messageLoggingInit().

void _checkpointBarrierAckHandler ( CheckpointBarrierMsg * msg )

Definition at line 3270 of file ckcausalmlog.C.

Referenced by _messageLoggingInit().

void garbageCollectMlogForChare	(	void *	data,
		ChareMlogData *	mlogData
	)

Function to remove all messages in the message log of a particular chare.

Definition at line 3289 of file ckcausalmlog.C.

References CkQ< T >::deq(), ChareMlogData::getMlog(), and CkQ< T >::length().

Referenced by garbageCollectMlog().

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 informLocationHome	(	CkGroupID	locMgrID,
		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.

Referenced by _distributedLocationHandler(), _receiveMigrationNoticeAckHandler(), _resendMessagesHandler(), CkLocMgr::informHome(), pupArrayElementsSkip(), and updateHomePE().

void _receiveLocationHandler ( CurrentLocationMsg * data )

Definition at line 3353 of file ckcausalmlog.C.

Referenced by _messageLoggingInit().

void _getGlobalStepHandler ( LBStepMsg * msg )

Definition at line 3394 of file ckcausalmlog.C.

Referenced by _messageLoggingInit().

void _recvGlobalStepHandler ( LBStepMsg * msg )

Receives the global step handler from PE 0.

Definition at line 3406 of file ckcausalmlog.C.

Referenced by _messageLoggingInit().

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.

Referenced by _bufferedLocalMessageCopyHandler(), _receiveMigrationNoticeHandler(), _removeProcessedLogHandler(), _restartHandler(), _updateHomeRequestHandler(), CkMlogRestart(), sendBufferedLocalMessageCopy(), sendLocalMessageCopy(), sendMlogLocation(), and startMlogCheckpoint().

envelope* copyEnvelope ( envelope * env )

Definition at line 3958 of file ckcausalmlog.C.

Referenced by generateCommonTicketRequest(), resendMessageForChare(), and sendCommonMsg().

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().

Variable Documentation

int _numRestartResponses = 0

Definition at line 56 of file ckcausalmlog.C.

int countHashRefs = 0

Definition at line 59 of file ckcausalmlog.C.

int countHashCollisions = 0

Definition at line 60 of file ckcausalmlog.C.

char* checkpointDirectory = "."

Definition at line 62 of file ckcausalmlog.C.

Referenced by readCheckpointFromDisk(), and writeCheckpointToDisk().

int unAckedCheckpoint = 0

Definition at line 63 of file ckcausalmlog.C.

int countLocal = 0

Definition at line 65 of file ckcausalmlog.C.

int countBuffered = 0

Definition at line 65 of file ckcausalmlog.C.

Referenced by sendBufferedLocalMessageCopy().

int countPiggy = 0

Definition at line 66 of file ckcausalmlog.C.

int countClearBufferedLocalCalls = 0

Definition at line 67 of file ckcausalmlog.C.

int countUpdateHomeAcks = 0

Definition at line 69 of file ckcausalmlog.C.

char* faultFile

Definition at line 77 of file ckcausalmlog.C.

Referenced by readFaultFile().

int faultFlag = 0

Definition at line 79 of file ckcausalmlog.C.

int restartingMlogFlag = 0

Definition at line 80 of file ckcausalmlog.C.

double killTime = 0.0

Definition at line 82 of file ckcausalmlog.C.

double faultMean

Definition at line 83 of file ckcausalmlog.C.

int checkpointCount = 0

Definition at line 84 of file ckcausalmlog.C.

int _numBufferedDets

Definition at line 110 of file ckcausalmlog.C.

int _indexBufferedDets

Definition at line 112 of file ckcausalmlog.C.

int _phaseBufferedDets

Definition at line 114 of file ckcausalmlog.C.

int _maxBufferedDets

Definition at line 119 of file ckcausalmlog.C.

int* numMsgsTarget

Definition at line 132 of file ckcausalmlog.C.

int* sizeMsgsTarget

Definition at line 133 of file ckcausalmlog.C.

int totalMsgsTarget

Definition at line 134 of file ckcausalmlog.C.

float totalMsgsSize

Definition at line 135 of file ckcausalmlog.C.

Referenced by _messageLoggingExit(), _messageLoggingInit(), and sendMsg().

int numPiggyDets

Definition at line 138 of file ckcausalmlog.C.

int numDets

Definition at line 139 of file ckcausalmlog.C.

int numDupDets

Definition at line 140 of file ckcausalmlog.C.

int msgLogSize

Definition at line 143 of file ckcausalmlog.C.

int bufferedDetsSize

Definition at line 144 of file ckcausalmlog.C.

int storedDetsSize

Definition at line 145 of file ckcausalmlog.C.

float MLOGFT_totalLogSize = 0.0

Definition at line 149 of file ckcausalmlog.C.

Referenced by _messageLoggingExit(), sendMsg(), and sendTicketRequest().

float MLOGFT_totalMessages = 0.0

Definition at line 150 of file ckcausalmlog.C.

Referenced by _messageLoggingExit(), sendMsg(), and sendTicketRequest().

double adjustChkptPeriod = 0.0 [static]

Definition at line 153 of file ckcausalmlog.C.

double nextCheckpointTime = 0.0 [static]

Definition at line 154 of file ckcausalmlog.C.

CkHashtableT<CkHashtableAdaptorT<CkObjID>,CkHashtableT<CkHashtableAdaptorT<CkObjID>,SNToTicket *> *> detTable(1000, 0.3) [static]

Referenced by populateDeterminantTable(), and removeProcessedLogs().

int _pingHandlerIdx

Definition at line 157 of file ckcausalmlog.C.

char objString[100]

Definition at line 159 of file ckcausalmlog.C.

Referenced by buildProcessedTicketLog(), createObjIDList(), and processReceivedTN().

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 _getCheckpointHandlerIdx

Definition at line 163 of file ckcausalmlog.C.

int _recvCheckpointHandlerIdx

Definition at line 164 of file ckcausalmlog.C.

int _removeProcessedLogHandlerIdx

Definition at line 165 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 _getGlobalStepHandlerIdx

Definition at line 172 of file ckcausalmlog.C.

int _recvGlobalStepHandlerIdx

Definition at line 173 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 _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 _storeDeterminantsHandlerIdx

Definition at line 183 of file ckcausalmlog.C.

int _removeDeterminantsHandlerIdx

Definition at line 184 of file ckcausalmlog.C.

int _restartHandlerIdx

Definition at line 187 of file ckcausalmlog.C.

int _getRestartCheckpointHandlerIdx

Definition at line 188 of file ckcausalmlog.C.

int _recvRestartCheckpointHandlerIdx

Definition at line 189 of file ckcausalmlog.C.

int verifyAckTotal

Definition at line 193 of file ckcausalmlog.C.

int verifyAckCount

Definition at line 194 of file ckcausalmlog.C.

int verifyAckedRequests = 0

Definition at line 196 of file ckcausalmlog.C.

RestartRequest* storedRequest

Definition at line 198 of file ckcausalmlog.C.

int _falseRestart = 0

Definition at line 200 of file ckcausalmlog.C.

int onGoingLoadBalancing = 0

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.

Referenced by _checkpointBarrierAckHandler(), and startLoadBalancingMlog().

void(* resumeLbFnPtr)(void *)

Referenced by startLoadBalancingMlog().

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 _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 countLBMigratedAway = 0

Definition at line 220 of file ckcausalmlog.C.

int countLBToMigrate = 0

Definition at line 221 of file ckcausalmlog.C.

int migrationDoneCalled = 0

Definition at line 222 of file ckcausalmlog.C.

int checkpointBarrierCount = 0

Definition at line 223 of file ckcausalmlog.C.

CkGroupID globalLBID

Definition at line 225 of file ckcausalmlog.C.

int restartDecisionNumber = -1

Definition at line 226 of file ckcausalmlog.C.

double lastCompletedAlarm = 0

Definition at line 228 of file ckcausalmlog.C.

double lastRestart = 0

Definition at line 229 of file ckcausalmlog.C.

int _receiveLocationHandlerIdx

Definition at line 232 of file ckcausalmlog.C.

int calledRetryTicketRequest = 0

This method is used to retry the ticket requests that had been queued up earlier.

Definition at line 1162 of file ckcausalmlog.C.

CkVec<TProcessedLog> processedTicketLog

Definition at line 1175 of file ckcausalmlog.C.

double totalSearchRestoredTime = 0

Definition at line 3762 of file ckcausalmlog.C.

double totalSearchRestoredCount = 0

Definition at line 3763 of file ckcausalmlog.C.


Data Structures
class	ElementPacker
class	ElementDistributor
class	ElementCaller
Functions
const char *	idx2str (const CkArrayIndex &ind)
const char *	idx2str (const ArrayElement *el)
void	getGlobalStep (CkGroupID gID)
bool	fault_aware (CkObjID &recver)
void	sendCheckpointData (int mode)
	Sends the checkpoint to its buddy.
void	createObjIDList (void data, ChareMlogData mlogData)
bool	isLocal (int destPE)
	Determines if the message is local or not.
bool	isTeamLocal (int destPE)
	Determines if the message is group local or not.
void	printLog (TProcessedLog *log)
	Prints a processed log.
void	readKillFile ()
	CpvDeclare (Chare *, _currentObj)
	CpvDeclare (StoredCheckpoint *, _storedCheckpointData)
	CpvDeclare (CkQ< MlogEntry * > *, _delayedLocalMsgs)
	CpvDeclare (Queue, _outOfOrderMessageQueue)
	CpvDeclare (Queue, _delayedRemoteMessageQueue)
	CpvDeclare (char **, _bufferedTicketRequests)
	CpvDeclare (int *, _numBufferedTicketRequests)
	CpvDeclare (char *, _localDets)
	CpvDeclare (CkDeterminantHashtableT *, _remoteDets)
	CpvDeclare (char *, _incarnation)
	CpvDeclare (CkVec< LocationID * > *, _emigrantRecObjs)
	CpvDeclare (CkVec< CkLocRec_local * > *, _immigrantRecObjs)
void	setTeamRecovery (void data, ChareMlogData mlogData)
	Turns on the flag for team recovery that selectively restores particular metadata information.
void	unsetTeamRecovery (void data, ChareMlogData mlogData)
	Turns off the flag for team recovery.
void	_messageLoggingInit ()
	Initialize message logging data structures and register handlers.
void	killLocal (void *_dummy, double curWallTime)
void	readFaultFile ()
	: reads the PE that will be failing throughout the execution and the mean time between failures.
void	addBufferedDeterminant (CkObjID sender, CkObjID receiver, MCount SN, MCount TN)
	Adds a determinants to the buffered determinants and checks whether the array of buffered determinants needs to be extended.
void	sendGroupMsg (envelope *env, int destPE, int _infoIdx)
	Sends a group message that might be a broadcast.
void	sendNodeGroupMsg (envelope *env, int destNode, int _infoIdx)
	Sends a nodegroup 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	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	_removeDeterminantsHandler (char *buffer)
	Removes the determinants after a particular index in the _localDets array.
void	_storeDeterminantsHandler (char *buffer)
	Stores the determinants coming from other processor.
void	_ticketRequestHandler (TicketRequest *ticketRequest)
	If there are any delayed requests, process them first before processing this request.
bool	_getTicket (envelope env, int flag)
	Gets a ticket for a recently received message.
int	preProcessReceivedMessage (envelope env, Chare objPointer, MlogEntry *logEntryPointer)
void	postProcessReceivedMessage (Chare obj, CkObjID &sender, MCount SN, MlogEntry entry)
	Updates a few variables once a message has been processed.
void	generalCldEnqueue (int destPE, envelope *env, int _infoIdx)
void	_pingHandler (CkPingMsg *msg)
void	buildProcessedTicketLog (void data, ChareMlogData mlogData)
	A chare adds the latest ticket number processed.
void	clearUpMigratedRetainedLists (int PE)
void	checkpointAlarm (void *_dummy, double curWallTime)
void	_checkpointRequestHandler (CheckpointRequest *request)
void	startMlogCheckpoint (void *_dummy, double curWallTime)
	Starts the checkpoint phase after migration.
void	pupArrayElementsSkip (PUP::er &p, CmiBool create, MigrationRecord *listToSkip, int listsize)
	Pups all the array elements in this processor.
void	writeCheckpointToDisk (int size, char *chkpt)
void	_storeCheckpointHandler (char *msg)
void	sendRemoveLogRequests ()
	Sends out the messages asking senders to throw away message logs below a certain ticket number.
void	_checkpointAckHandler (CheckPointAck *ackMsg)
void	populateDeterminantTable (char *data)
	Inserts all the determinants into a hash table.
void	removeProcessedLogs (void _data, ChareMlogData mlogData)
void	_removeProcessedLogHandler (char *requestMsg)
	Removes messages in the log according to the received ticket numbers.
void	CkMlogRestart (const char dummy, CkArgMsg dummyMsg)
	Function for restarting the crashed processor.
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	CkMlogRestartDouble (void *, double)
void	CkMlogRestartLocal ()
void	_getCheckpointHandler (RestartRequest *restartMsg)
	Gets the stored checkpoint for its buddy processor.
void	_verifyAckRequestHandler (VerifyAckMsg *verifyRequest)
void	_verifyAckHandler (VerifyAckMsg *verifyReply)
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	_updateHomeAckHandler (RestartRequest *updateHomeAck)
	Receives the updateHome ACKs from all other processors.
void	initializeRestart (void data, ChareMlogData mlogData)
	Initializes variables and flags for restarting procedure.
void	updateHomePE (void data, ChareMlogData mlogData)
	Updates the homePe of chare array elements.
void	_updateHomeRequestHandler (RestartRequest *updateRequest)
	Updates the homePe for all chares in this processor.
void	fillTicketForChare (void data, ChareMlogData mlogData)
	Fills up the ticket vector for each chare.
void	printMsg (envelope env, const char par)
	Prints information about a message.
void	printDet (Determinant det, const char par)
	Prints information about a determinant.
void	resendMessageForChare (void data, ChareMlogData mlogData)
	Resends all the logged messages to a particular chare list.
void	_sendDetsHandler (char *msg)
	Send all remote determinants to a particular failed PE.
void	_resendMessagesHandler (char *msg)
	Resends messages since last checkpoint to the list of objects included in the request.
MCount	maxVec (CkVec< MCount > *TNvec)
	Returns the maximum ticket from a vector.
void	sortVec (CkVec< MCount > *TNvec)
int	searchVec (CkVec< MCount > *TNVec, MCount searchTN)
void	processDelayedRemoteMsgQueue ()
	Processes the messages in the delayed remote message queue.
void	_sendDetsReplyHandler (char *msg)
	Receives determinants stored on remote nodes.
void	_receivedDetDataHandler (ReceivedDetData *msg)
	Receives a list of determinants coming from the home PE of a migrated object (parallel restart).
void	_receivedTNDataHandler (ReceivedTNData *msg)
	Receives a list of TNs coming from the home PE of a migrated object (parallel restart).
void	processReceivedDet (Chare obj, int listSize, Determinant listDets)
	Processes the received list of determinants from a particular PE.
void	processReceivedTN (Chare obj, int listSize, MCount listTNs)
	Processes the received list of tickets from a particular PE.
void	distributeRestartedObjects ()
	Distributes objects to accelerate recovery after a failure.
void	_distributedLocationHandler (char *receivedMsg)
	Handler to update information about an object just received.
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	sendDummyMigrationCounts (int *dummyCounts)
	this method is used by a restarted processor to tell other processors that they are not going to receive these many objects.
void	_dummyMigrationHandler (DummyMigrationMsg *msg)
	this handler is used to process a dummy migration msg.
void	forAllCharesDo (MlogFn fnPointer, void *data)
	Map function pointed by fnPointer over all the chares living in this processor.
void	initMlogLBStep (CkGroupID gid)
	This is the first time Converse is called after AtSync method has been called by every local object.
void	startLoadBalancingMlog (void(_fnPtr)(void ), void *_centralLb)
	Load Balancing.
void	finishedCheckpointLoadBalancing ()
void	sendMlogLocation (int targetPE, envelope *env)
void	_receiveMigrationNoticeHandler (MigrationNotice *msg)
void	_receiveMigrationNoticeAckHandler (MigrationNoticeAck *msg)
void	_receiveMlogLocationHandler (void *buf)
void	resumeFromSyncRestart (void data, ChareMlogData mlogData)
void	checkAndSendCheckpointBarrierAcks (CheckpointBarrierMsg *msg)
	Processor 0 sends a broadcast to every other processor after checkpoint barrier.
void	_checkpointBarrierHandler (CheckpointBarrierMsg *msg)
	Processor 0 receives a contribution from every other processor after checkpoint.
void	_checkpointBarrierAckHandler (CheckpointBarrierMsg *msg)
void	garbageCollectMlogForChare (void data, ChareMlogData mlogData)
	Function to remove all messages in the message log of a particular chare.
void	garbageCollectMlog ()
	Garbage collects the message log and other data structures.
void	informLocationHome (CkGroupID locMgrID, 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	_getGlobalStepHandler (LBStepMsg *msg)
void	_recvGlobalStepHandler (LBStepMsg *msg)
	Receives the global step handler from PE 0.
void	_messageLoggingExit ()
	Function to wrap up performance information.
int	getCheckPointPE ()
	Getting the pe number of the current processor's buddy.
envelope *	copyEnvelope (envelope *env)
int	isSameDet (Determinant first, Determinant second)
Variables
int	_restartFlag = 0
int	_numRestartResponses = 0
int	countHashRefs = 0
int	countHashCollisions = 0
char *	checkpointDirectory = "."
int	unAckedCheckpoint = 0
int	countLocal = 0
int	countBuffered = 0
int	countPiggy = 0
int	countClearBufferedLocalCalls = 0
int	countUpdateHomeAcks = 0
int	teamSize
int	chkptPeriod
bool	fastRecovery
int	parallelRecovery
char *	killFile
char *	faultFile
int	killFlag = 0
int	faultFlag = 0
int	restartingMlogFlag = 0
double	killTime = 0.0
double	faultMean
int	checkpointCount = 0
int	_numBufferedDets
int	_indexBufferedDets
int	_phaseBufferedDets
int	_maxBufferedDets
int *	numMsgsTarget
int *	sizeMsgsTarget
int	totalMsgsTarget
float	totalMsgsSize
int	numPiggyDets
int	numDets
int	numDupDets
int	msgLogSize
int	bufferedDetsSize
int	storedDetsSize
float	MLOGFT_totalLogSize = 0.0
float	MLOGFT_totalMessages = 0.0
static double	adjustChkptPeriod = 0.0
static double	nextCheckpointTime = 0.0
static CkHashtableT < CkHashtableAdaptorT< CkObjID > , CkHashtableT < CkHashtableAdaptorT< CkObjID > , SNToTicket * > * >	detTable (1000, 0.3)
int	_pingHandlerIdx
char	objString [100]
int	_checkpointRequestHandlerIdx
int	_storeCheckpointHandlerIdx
int	_checkpointAckHandlerIdx
int	_getCheckpointHandlerIdx
int	_recvCheckpointHandlerIdx
int	_removeProcessedLogHandlerIdx
int	_verifyAckRequestHandlerIdx
int	_verifyAckHandlerIdx
int	_dummyMigrationHandlerIdx
int	_getGlobalStepHandlerIdx
int	_recvGlobalStepHandlerIdx
int	_updateHomeRequestHandlerIdx
int	_updateHomeAckHandlerIdx
int	_resendMessagesHandlerIdx
int	_sendDetsHandlerIdx
int	_sendDetsReplyHandlerIdx
int	_receivedTNDataHandlerIdx
int	_receivedDetDataHandlerIdx
int	_distributedLocationHandlerIdx
int	_storeDeterminantsHandlerIdx
int	_removeDeterminantsHandlerIdx
int	_restartHandlerIdx
int	_getRestartCheckpointHandlerIdx
int	_recvRestartCheckpointHandlerIdx
int	verifyAckTotal
int	verifyAckCount
int	verifyAckedRequests = 0
RestartRequest *	storedRequest
int	_falseRestart = 0
int	onGoingLoadBalancing = 0
	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	_checkpointBarrierHandlerIdx
int	_checkpointBarrierAckHandlerIdx
CkVec< MigrationRecord >	migratedNoticeList
CkVec< RetainedMigratedObject * >	retainedObjectList
int	donotCountMigration = 0
int	countLBMigratedAway = 0
int	countLBToMigrate = 0
int	migrationDoneCalled = 0
int	checkpointBarrierCount = 0
int	globalResumeCount = 0
CkGroupID	globalLBID
int	restartDecisionNumber = -1
double	lastCompletedAlarm = 0
double	lastRestart = 0
int	_receiveLocationHandlerIdx
int	calledRetryTicketRequest = 0
	This method is used to retry the ticket requests that had been queued up earlier.
CkVec< TProcessedLog >	processedTicketLog
double	totalSearchRestoredTime = 0
double	totalSearchRestoredCount = 0