(19)
(11)EP 2 756 406 B1

(12)EUROPEAN PATENT SPECIFICATION

(45)Mention of the grant of the patent:
22.07.2020 Bulletin 2020/30

(21)Application number: 11872383.2

(22)Date of filing:  14.09.2011
(51)International Patent Classification (IPC): 
G06F 12/16(2006.01)
G06F 9/46(2006.01)
G06F 12/00(2006.01)
(86)International application number:
PCT/US2011/051516
(87)International publication number:
WO 2013/039494 (21.03.2013 Gazette  2013/12)

(54)

IMPARTING DURABILITY TO A TRANSACTIONAL MEMORY SYSTEM

HERSTELLUNG DER HALTBARKEIT EINES TRANSAKTIONSSPEICHERSYSTEMS

PROCÉDÉ POUR CONFÉRER UNE DURABILITÉ À UN SYSTÈME DE MÉMOIRE DE TRANSACTIONS


(84)Designated Contracting States:
AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MK MT NL NO PL PT RO RS SE SI SK SM TR

(43)Date of publication of application:
23.07.2014 Bulletin 2014/30

(73)Proprietor: Hewlett Packard Enterprise Development LP
Houston, TX 77070 (US)

(72)Inventors:
  • SHAH, Mehul A.
    Palo Alto, California 94304-1100 (US)
  • HARIZOPOULOS, Stavros
    Palo Alto, California 94304-1100 (US)
  • MERCHANT, Arif A.
    Los Altos, California 94022 (US)
  • SAXENA, Mohit
    Madison, Wisconsin 53715-1173 (US)

(74)Representative: Fleuchaus, Michael A. et al
Fleuchaus & Gallo Partnerschaft mbB Patentanwälte Steinerstraße 15/Haus A
81369 München
81369 München (DE)


(56)References cited: : 
US-A- 5 913 219
US-A1- 2007 011 416
US-A1- 2009 193 193
US-A1- 2011 072 207
US-A1- 2004 064 635
US-A1- 2009 150 599
US-A1- 2009 217 018
  
      
    Note: Within nine months from the publication of the mention of the grant of the European patent, any person may give notice to the European Patent Office of opposition to the European patent granted. Notice of opposition shall be filed in a written reasoned statement. It shall not be deemed to have been filed until the opposition fee has been paid. (Art. 99(1) European Patent Convention).


    Description

    Background



    [0001] A transactional memory system allows application programmers to safely and easily scale their programs across many processors or processing cores. In a typical transactional memory system, multiple application threads share a volatile memory, such as a dynamic random access memory (DRAM), while isolation is provided among the threads for accessing the volatile memory. A transactional memory system typically is faster than a traditional transactional database system that employs the use of mass storage that has relatively slow access times, such as magnetic storage, rotating disks, tape drives, etc.

    [0002] US2009/150599 (Jon C.R. Bennett) discloses managing the storage of transaction data in volatile and non-volatile memory.

    Brief Description Of The Drawing



    [0003] 

    Fig. 1 is a layer diagram of a physical machine according to an example implementation.

    Fig. 2 is a flow diagram depicting a technique to impart durability to a transactional memory system according to an example implementation.

    Fig. 3 is an illustration of an architecture to impart durability to a transactional memory system according to an example implementation.

    Fig. 4 is a flow diagram depicting a technique to log transaction records according to an example implementation.

    Fig. 5 is a flow diagram depicting a technique to create a checkpoint according to an example implementation.


    Detailed Description



    [0004] Systems and techniques are disclosed herein for purposes of imparting durability to a transactional memory system, which relies on a volatile memory (a dynamic random access memory (DRAM), for example) as its primary storage. In this context, a "transactional memory system" is a system that allows multiple processing entities (threads, for example) to share a memory while providing isolation among the threads for accesses to the memory. The transactional memory system allows a given processing entity to perform a transaction that logically appears to the other processing entities to occur at a single point in time, while in reality the transaction may involve a sequence of write operations, read operations and interspersed computations by the given processing entity.

    [0005] As a non-limiting example, a given processing entity may perform a transaction for purposes of modifying a data structure in the memory, which involves reading parameters of the structure, modifying parameters of the structure and writing back resulting modified parameters to the memory. The transactional memory system regulates the transactions so that a single transaction at a time occurs with the memory, although, as noted above, the transaction may involve multiple memory operations, and some of these memory operations may occur in parallel with the memory. The results of a given transaction for a given processing entity become visible to the other processing entities after the transactional memory system commits the transaction to the memory.

    [0006] The transactional memory system may either be software-based (e.g., implemented using software that operates with the operating system, software that is part of the operating system, software entirely separate from the operation system) or hardware-based (e.g. implemented using hardware inside a microprocessor, as a non-limiting example), depending on the particular implementation.

    [0007] Regardless of its particular form, the transactional memory system instills the properties of atomicity, consistency and isolation for the transactions. Referring to Fig. 1, in accordance with example implementations disclosed herein, a durable transactional memory system 39 includes a transactional memory system 40 (which may be a software or hardware transactional memory system) that has these three properties and a transaction durability engine 50, which imparts the additional property of durability for the transactions, even though the transactional memory system 39 relies on a volatile memory 36 as primary storage for the transactions. As further described below, although the volatile memory 36 is potentially subject to an event (an event in which power to the volatile memory 36 is lost, for example) that may erase or undesirably alter the contents of the volatile memory 36, the transaction durability engine 50 selectively stores data in a non-volatile memory 37 to ensure that a consistent state of the transactional memory system 40 may be recovered if such an event occurs.

    [0008] As depicted in Fig. 1, in accordance with some example implementations, the durable transactional memory system 39, volatile memory 36 and non-volatile memory 37 may be part of a physical machine 10. In this context, a "physical machine" indicates that the machine is an actual machine made up of executable program instructions and hardware. Examples of physical machines include computers (e.g., application servers, storage servers, web servers, etc.), communications modules (e.g., switches, routers, etc.) and other types of machines. The physical machine 10 may be located within one cabinet (or rack); or alternatively, the physical machine 10 may be located in multiple cabinets (or racks).

    [0009] The physical machine 10 may be implemented in an application server, a storage server farm (or storage area network), a web server farm, a switch or router farm, other type of data center, and so forth. Also, although the physical machine 10 is depicted in Fig. 1 as being contained within a box, it is noted that the physical machine 10 may be a distributed machine having multiple nodes, which provide a distributed and parallel processing system.

    [0010] In accordance with a specific example described herein, the physical machine 10 contains machine executable instructions, which are executed by hardware of the machine 10, such as one or multiple processors 34. In general, the processors 34 may be one or multiple central processing units (CPUs) and/or may be one or multiple processing cores 34 of one or more CPUs.

    [0011] As a non-limiting example of the machine executable instructions, the physical machine 10 may contain a set of instructions that form an operating system 28, as well as sets of machine executable instructions that when executed form respective applications 26 and drivers 30. In accordance with other example implementations, the physical machine 10 may not include the operating system 28 and/or may not include the drivers 30. In accordance with an example implementation, a set of machine executable instructions may when executed form the transactional memory system 40 and a set of machine executable instructions may when executed form the durable transactional memory system 39, including the durability engine 50 and transactional memory system 40, although any of these components may be implemented in hardware in accordance with other example implementations.

    [0012] Among its other features, the physical machine 10 may include additional hardware, such as at least one cache 35 as well as solid state drives (SSDs) 31 that are operated via one or more of the drivers 30 and collectively form the non-volatile memory 37, in accordance with some example implementations. As a non-limiting example, a given SSD 31 may contain NAND flash memory, although the SSD 31 may store data in other non-volatile storage mediums, such as phase change memory (PCM) or memristor memory, in accordance with other example implementations.

    [0013] Referring to Fig. 2 in conjunction with Fig. 1, to summarize, the transaction durability engine 50, when executed, causes the physical machine 10 to perform a technique 150 for purposes of imparting durability to the durable transactional memory system 39. Pursuant to the technique 150, the physical machine 10 contains a transactional memory system that uses (block 154) volatile memory 36 as its primary storage; and the physical machine 10 selectively stores (block 158) data in a non-volatile memory 37 to impart durability to the durable transactional memory system 39 to allow the transactional memory system 40 to be restored to a consistent state in the event of data loss to the volatile memory 36.

    [0014] As a more detailed example, Fig. 3 depicts an architecture used by the transaction durability engine 50 according to an example implementation. Referring to Fig. 1 in conjunction with Fig. 3, for this example, the durable transactional memory system 39 employs the use of a memory heap 200 in volatile memory storage 199 (storage created by the volatile memory 36, for example) as its primary storage. In other words, the memory heap 200 contains the full record of the current state of the transactional memory system 40 and application state, and as further described below, is divided by the transaction durability engine 50 into chunks 204.

    [0015] As also depicted in Figs. 1 and 3, the transaction durability engine 50 stores its data in non-volatile memory storage 220 (storage in the non-volatile memory 37, for example). More specifically, the transaction durability engine 50 stores data in the non-volatile memory storage 220 through the use of partitioned logging (i.e., records stored in log partitions 260) and checkpoints 280 to ensure that the durable transactional memory system 39 may be restored to a consistent state, in the event of that the volatile memory storage 199 experiences a data loss. The log partition 260 is, in general, a data structure in the non-volatile memory storage 220, which is associated with a particular processing entity and in which records that identify corresponding transactions are written. The checkpoint 280, in general, is a full record of the state of the volatile memory heap 200 at some point or range of points in time, such that the checkpoint 280, in conjunction with the logged transaction information provided by the records stored in the log partitions 260, may be used to recover a consistent state of the durable transactional memory system 39.

    [0016] The durable transactional memory system 39 and the transaction durability engine 50 both rely on the use of logical sequence numbers (LSNs), in accordance with example implementations. The LSN defines the order of an associated transaction and is consistent with the partial order of the transactions. Collectively, the LSNs define the order in which the transactions are committed. As a non-limiting example, the LSNs may be sequential integers. However, in accordance with other example implementations, the LSNs may be any set of ordered values, in accordance with the many potential implementations.

    [0017] In general, processing entities (transaction threads 202 for example) generate transactions (a given transaction being a sequence of write operations, read operations and interspersed computations, for example), and the transactional memory system 40 assigns an LSN to each transaction. The transaction durability engine 50 does not re-assign the LSNs but rather designates the LSNs in different ways, depending on how the LSN is being used. For example, as further described below, the transaction durability engine 50 designates some of these LSNs as "logger LSNs 255," which indicate the state of the log and partitioned logs, respectively, as further described below. The logger LSN is the LSN of the most recent transaction from the log queue 252 that has been written to a log partition 260 in the non-volatile memory 37. The transaction durability engine 50 also employs the use a "committed LSN," which is the minimum of all logger LSN 255 values and is the LSN of the transaction for which all preceding transaction records 254 have been written out to the non-volatile memory storage 220.

    [0018] The transaction durability engine 50 also employs the use of other LSNs. In this manner, a "chunk LSN" 205 refers to the LSN of the last transaction that checkpointed an associated chunk 204. A "start LSN" 290 indicates that all previous transactions in a valid checkpoint 280 are reflected in the checkpoint 280. An "end LSN" 294 indicates the point where subsequent transactions are definitely not reflected in the checkpoint 280.

    [0019] Turning now to more specific details, in accordance with some implementations, the transaction durability engine 50 includes a logger 54 (see Fig. 1), which causes the physical machine 10 to create a corresponding logger thread 250 for each processing entity that creates transactions. For the following non-limiting example, it is assumed that a logger thread is created for each transaction thread 202, and the transaction threads 202 may be associated with the execution(s) of one or multiple applications 26. After a given transaction thread 202 has completed its heap 200 updates, the transaction thread 202 enqueues a corresponding transaction record 254 of the updates it has made into an associated log queue 252 that is stored in the volatile memory storage 199. In general, each log queue 252 is associated with a given transaction thread 202 and log partition 260 (for this example); and each log queue 252 stores transaction records 254 that reflect the associated transactions that were committed to the heap 200.

    [0020] Periodically, the logger thread 250 reports the committed LSN value to the transactional threads 202, which means that all transactions with LSN values less than or equal to the committed LSN value have been saved to the non-volatile memory region 220. The rate at which the committed LSN computation is made may be based on a number of various factors, such as a combination of a time threshold and the completion of a set of transaction log writes. Moreover, the transaction thread 202 informs the calling applications 26 that the corresponding transactions have been committed. It is noted that the applications 26 may or may not wait for this confirmation, depending on whether the durable transactional memory system 39 is in a synchronous or asynchronous mode operation.

    [0021] More specifically, when the durable transactional memory system 39 is in the synchronous mode of operation, the system 39 confirms to a calling application or thread that a given transaction has been committed after the transaction has been stored in the volatile 36 and non-volatile 37 memories. When the durable transactional memory system 39 is in the asynchronous mode of operation, the system 39 confirms to a calling application 26 or thread that a given transaction has been committed after the transaction has been stored in the volatile memory 36 but before the transaction has been stored in the non-volatile memory 37. Therefore, except for an event that compromises the integrity of the volatile memory 36, such as a power failure (as a non-limiting example), consistency is preserved, without incurring the delay that would be experienced by a truly synchronous system. In accordance with some implementations, in addition, the durable transactional memory system 39 provides an interface for an application 26 or thread to determine the durability of a given asynchronous transaction (i.e., the application 26/thread may use the interface to determine whether a given asynchronous transaction has been stored in the non-volatile memory 37).

    [0022] In some example implementations, the logger thread 250 periodically writes out of set of transaction records 254 from the log queue 252 into a corresponding log partition 260 in a first-in first-out (FIFO) order, depending on such factors as whether the size of the queued records exceeds a size threshold or whether a certain time interval has elapsed after the records were last updated to the log partition 260. In accordance with some example implementations, the non-volatile memory storage 220 enqueues multiple such write requests from multiple logger threads 250 and writes these to update more than one log partition 260, simultaneously, or in parallel. When the new records are stored in the non-volatile memory storage 220, the non-volatile memory 37 informs the appropriate logger thread(s) 250; and the logger thread(s) 250 then update the corresponding logger LSN(s) 255 to reflect the durable LSNs for each log queue 252.

    [0023] Thus, referring to Fig. 4 in conjunction with Fig. 3, in accordance with some example implementations, the logger thread 250 performs a technique 300. Pursuant to the technique 300, transaction records are received into the log queue 252, pursuant to block 304. The logger thread 250 determines (diamond 308) whether it is time to write records from the log queue 252 to the log partition 260 in the non-volatile memory storage 220. If so, then the logger thread 250 writes the records to the log partition 260, pursuant to block 312 and upon completion of the write operations, updates the logger LSN, pursuant to block 314.

    [0024] Control then proceeds to diamond 318, where the logger thread 250 determines (diamond 318) whether it is time to report a committed LSN. If so, then the logger thread 250 communicates the committed LSN to the corresponding transaction thread(s) 202, pursuant to block 322. Control then returns to block 304.

    [0025] As a non-limiting example, the logger thread 250 may determine the committed LSN every time that a log write completes at the non-volatile memory storage 220 or if the logger thread 250 has corresponding pending log writes with an LSN greater than the last reported committed LSN and the time since the previous check exceeds a set time threshold. Alternatively, in accordance with other example implementations, the transaction memory manager 50 may cause the physical machine 10 to create a thread separate from the logger threads 250 for purposes of periodically determining the committed LSN. The communication of the newly committed LSN in either case may be reported it to all of the transaction threads 202, either through (as non-limiting examples) a message or by posting the committed LSN to a common location, which the transaction threads 202 may access via polling.

    [0026] Referring to Fig. 1 in conjunction with Fig. 3, for purposes of creating the checkpoint 280, the transaction durability engine 50 includes a checkpointer 52, which when executed causes the physical machine 10 to create a corresponding checkpointer thread 270. As noted above, the checkpoint 280 is a full record of the state of the memory heap 200 at some point or range of points in time, such that the checkpoint 280, in conjunction with the records logged in the log partitions 260, may be used to recover the most recent consistent state of the durable transactional memory system 39.

    [0027] It is noted that in accordance with some example implementations, the checkpoint 280 may not be a snapshot of the memory heap 200 at a given single time. Instead, in accordance with some example implementations, the checkpoint 280 reflects the chunks 204 over a range of times. In general, the checkpoint 280 is stored in the non-volatile memory storage 220 and has an associated start LSN 290, an associated end LSN 294 and may also include an associated valid flag 296.

    [0028] The checkpoint 280 reflects all of the transactions with LSN less than or equal to the start LSN 290; and the checkpoint 280 reflects no transactions with an LSN greater than the end LSN 294. It is noted that the checkpoint 280 may not reflect all of the transactions with LSNs between the start LSN 290 and the end LSN 294.

    [0029] In general, a given checkpoint 280 is valid if all records 254 less than or equal to the end LSN 294 are written to the log partitions 260 and all records 254 greater than the start LSN 290 remain in the log partitions 260. All records 254 that are less than or equal to the start LSN 290 may be discarded. In general, there may be multiple checkpoints 280 stored in the non-volatile memory storage 220, with always at least one valid checkpoint 280.

    [0030] The checkpoint 280 may be created as follows, in accordance with some example implementations. At least one copy of the memory heap 200 exists before transactions may be run. For example, a checkpoint initialization process may simply copy the memory heap 200 to the non-volatile memory storage 220. Since no transactions have been executed, the start 290 and end 294 LSNs each have an initial value (zero and infinity, respectively, for example). After that, transactions may begin, and the checkpointer thread 270 creates a new checkpoint 280 as follows. At the beginning of the new checkpoint creation process, the checkpointer thread 270 creates an empty checkpoint stub in the non-volatile memory storage 220, with the start LSN 290 having its initial value (zero, for example), the end LSN 294 having an initial value (a maximum number or a number indicative of infinity, as non-limiting examples), and the valid flag 296 being false.

    [0031] In the creation of a new checkpoint 280, a chunk 204 of the memory heap 200 is copied to a temporary buffer using a transaction, and the LSN of that transaction is set as the chunk LSN 205 for that chunk 204. The content of the temporary buffer is then copied to the non-volatile memory 37 to create a copy of the chunk 204 in the new checkpoint 280. The first chunk 204 sets the start LSN 290, and the last chunk 204 sets the end LSN 294. As the chunks 204 are copied and written to the non-volatile memory storage 220, the end LSN 294 is updated to the latest chunk LSN 204.

    [0032] At the end of the copying process, the checkpointer thread 270 determines whether the checkpoint 280 is valid by verifying that all the logger LSN 255 values are equal to or greater than the determined end LSN 294, which indicates that all the updates reflected in the checkpoint 280 have been saved to a log. If not, the checkpointer thread 270 pauses until this condition no longer holds (which eventually becomes true if the logger threads 250 are not blocked). Alternatively, the checkpointer thread 270 may force the logger threads 250 to write enough of the loq queues to the log partitions 260 to ensure this conditions holds. Subsequently, the checkpointer thread 270 sets the valid flag 296 to a true state.

    [0033] In accordance with example implementations, the checkpointer thread 270 may remove one or multiple old checkpoints 280 or may reuse the memory space occupied by the older checkpoint(s) 280. Log records that are older than the valid checkpoint's start LSN may also be discarded, because these records are reflected in the newly-created checkpoint 280, in accordance with some example implementations. Because older checkpoint(s) 280 are removed after the newest checkpoint 280 is verified to be valid, there is at least one valid checkpoint 280 stored in the non-volatile memory storage 220 at all times.

    [0034] To summarize, referring to Fig. 5 in conjunction with Fig. 3, in accordance with an example implementation, the checkpointer thread 270 may perform a technique 350. Pursuant to the technique 350, the checkpointer thread 270 initializes (block 354) the LSN start, LSN end and valid flags in the non-volatile memory storage 220 and volatile memory storage 199 for a new checkpoint 280 before copying (block 358) the chunks 204 in the memory heap 200 to the new checkpoint 280, pursuant to block 358. Subsequently, the checkpointer thread 270 updates the end LSN 294 and start LSN 290 on the first chunk in memory, pursuant to block 362.

    [0035] In response to the checkpointer thread 270 determining (diamond 366) that more copying is needed, chunks 204 are continued to be copied, pursuant to block 358. Further, if more copying is to occur, the end LSN (and not the start LSN) is updated for subsequent chunks after the first, pursuant to block 362. It is noted that Fig. 5 merely schematically depicts the copying of the data for the new checkpoint 280, as the actually copying may occur in a predetermined number of iterations in a loop, may involve parallel copying, and so forth. After the copying is complete, the checkpointer thread 270 determines (diamond 370) whether the checkpoint is valid. If so, the checkpointer thread 270 sets the valid flag to true in the non-volatile memory storage 220 and updates the start and end LSNs in the non-volatile memory storage 220, pursuant to block 374.

    [0036] Referring back to Figs. 1 and 3, in accordance with some example implementations, transaction durability engine 50 includes another component, a recovery builder 58, for purposes of recovering stored transactional memory data should a failure occur (e.g., data loss due to a power failure). In this recovery, the recovery builder 58, when executed by the physical machine 10, restores a consistent state for the durable transactional memory system 39. In general, the recovery builder 58 uses the newest valid checkpoint 280 and applies the logged records in LSN order. In this manner, if the LSN of a given record is less than the chunk LSN, then the log record may be ignored for that chunk 204. After the rebuild, the memory heap 200 reflects all committed transactions. If no valid flag was stored for the checkpoints 280, the recovery builder 58 determines the validity of the checkpoints 280 by determining if the logger LSN 255 for each logger is not less than the end LSN of the checkpoint 280. The logger LSNs may be found as the last LSN of the log partition 260, because each logger thread 250 writes transaction records in the order of their LSNs.

    [0037] While a limited number of examples have been disclosed herein, those skilled in the art, having the benefit of this disclosure, will appreciate numerous modifications and variations therefrom. It is intended that the appended claims cover all such modifications and variations.


    Claims

    1. A method of implementing a transactional memory system (40) comprising:

    using a volatile memory (36) as primary storage for transactions associated with the transactional memory system (40); and

    selectively storing data in a non-volatile memory (37) to impart durability to the transactional memory system (40) to allow the transactional memory system (40) to be restored to a consistent state in the event of data loss to the volatile memory (36), wherein,

    the act of selectively storing comprises:

    copying data stored in the volatile memory (36) into a first non-volatile memory (37) region to generate a checkpoint (52);

    logging records (54) that reflect transactions that occurred after the generation of the checkpoint (52) in a second non-volatile memory (37) region;

    associating the stored data with ordering identifiers assigned by the transactional memory system (40) to transactions that generated the data; and

    providing a first indication to a thread that data associated with a transaction has been stored in the volatile memory (36) but before the data has been stored in the non-volatile memory (37) while the system is in an asynchronous mode and a second indication to the thread that the data has been stored in both the volatile memory (36) and the non-volatile memory (37) while the system is in a synchronous mode.


     
    2. The method of claim 1, wherein the logging records (54) of transactions processed by the transactional memory system (40) are logged in the non-volatile memory (37) in partitions, each of the partitions being associated with a different transaction generating entity.
     
    3. The method of claim 1, wherein the checkpoint (52) in the non-volatile memory (37) is indicative of a full record of a heap in the volatile memory (36) used by the transactional memory system.
     
    4. The method of claim 1, wherein the act of selectively storing comprises selectively storing the data in at least one of the following memories: NAND flash memory, phase change memory and memristor memory.
     
    5. The method of claim 1, further comprising, in response to a data loss in the volatile memory, reconstructing transactional memory data.
     
    6. An article comprising at least one machine-readable storage medium storing instructions that upon execution cause a system having at least one processor to:
    use a volatile memory (36) as primary storage for transactions associated with a transactional memory system; and selectively store data in a non-volatile memory (37) to impart durability to the transactional memory system (40) to allow the transactional memory system (40) to be restored to a consistent state in the event of data loss to the volatile memory; the machine-readable storage medium storing instructions that upon execution cause the system having the at least one processor further to:

    copy data stored in the volatile memory (36) into a first non-volatile memory (37) region to generate a checkpoint (52) and logging records (54) that reflect transactions that occurred after the generation of the checkpoint (52) in a second non-volatile memory (37) region;

    associate the stored data with ordering identifiers assigned by the transactional memory system (40) to transactions that generated the data;

    provide first confirmations to the threads indicating that asynchronous transactions have been stored in the volatile memory (36) to operate the system in an asynchronous mode and second confirmations to the threads indicating that asynchronous transactions have been stored in the non-volatile memory (37) to operate the system in a synchronous mode.


     
    7. An apparatus comprising: a volatile memory (36) to store transactional memory data created by a plurality of threads generating transactions with the volatile memory;
    a non-volatile memory; and
    a transactional memory system (40) adapted to:

    use the volatile memory (36) as primary storage for the transactions; and

    selectively store data in the non-volatile memory to impart durability to the transactional memory system (40) to allow the transactional memory system (40) to be restored to a consistent state in the event of data loss to the volatile memory; wherein

    the transactional memory system (40) is further to:

    copy data stored in the volatile memory (36) into a first non-volatile memory region to generate a checkpoint (52) and logging records (54) that reflect transactions that occurred after the generation of the checkpoint (52) in a second non-volatile memory region;

    associate the stored data with ordering identifiers assigned by the transactional memory system (40) to transactions that generated the data; and

    provide first confirmations to the threads indicating that asynchronous transactions have been stored in the volatile memory (36) to operate the system in an asynchronous mode and second confirmations to the threads indicating that asynchronous transactions have been stored in the non-volatile memory (37) to operate the system in a synchronous mode.


     
    8. The apparatus of claim 7, wherein the transactional memory system (40) is further adapted to log the logging records (54) of transactions processed by the transactional memory system (40) in the non-volatile memory (37) in partitions, each of the partitions being associated with a different transaction generating entity.
     
    9. The apparatus of claim 7, wherein the transactional memory system (40) is further adapted to store the checkpoint (52) in the non-volatile memory (37) indicative of a full record of a heap in the volatile memory (36) used by the transactional memory system(40).
     
    10. The apparatus of claim 7, wherein the non-volatile memory (37) comprises at least one of the following memories: NAND flash memory, phase change memory and memristor memory.
     
    11. The apparatus of claim 7, wherein the transactional memory system (40) is further adapted to reconstruct transactional memory data in response to a data loss in the volatile memory.
     


    Ansprüche

    1. Verfahren zum Implementieren eines Transaktionsspeichersystems (40), das Folgendes umfasst:

    Verwenden eines flüchtigen Speichers (36) als Primärspeicher für Transaktionen, die dem Transaktionsspeichersystem (40) zugeordnet sind; und

    selektives Speichern von Daten in einem nichtflüchtigen Speicher (37), um dem Transaktionsspeichersystem (40) Dauerhaftigkeit zu verleihen, um es dem Transaktionsspeichersystem (40) zu ermöglichen, im Falle eines Datenverlusts des flüchtigen Speichers (36) in einen konsistenten Zustand zurückversetzt zu werden, wobei der Vorgang des selektiven Speicherns Folgendes umfasst:

    Kopieren von in dem flüchtigen Speicher (36) gespeicherten Daten in einen ersten nichtflüchtigen Speicher(37)bereich, um einen Fixpunkt (52) zu erzeugen;

    Protokollieren von Datensätzen (54), die Transaktionen widerspiegeln, die nach der Erzeugung des Fixpunkts (52) in einem zweiten nichtflüchtigen Speicher(37)bereich aufgetreten sind;

    Zuordnen der gespeicherten Daten zu Ordnungskennungen, die von dem Transaktionsspeichersystem (40) Transaktionen zugewiesen werden, die die Daten erzeugt haben; und

    Bereitstellen einer ersten Angabe an einen Thread, dass Daten, die einer Transaktion zugeordnet sind, in dem flüchtigen Speicher (36) gespeichert wurden, jedoch bevor die Daten in dem nichtflüchtigen Speicher (37) gespeichert wurden, während sich das System in einem asynchronen Modus befindet, und einer zweiten Angabe an den Thread, dass die Daten sowohl in dem flüchtigen Speicher (36) als auch in dem nichtflüchtigen Speicher (37) gespeichert wurden, während sich das System in einem synchronen Modus befindet.


     
    2. Verfahren nach Anspruch 1, wobei die Protokolldatensätze (54) von Transaktionen, die von dem Transaktionsspeichersystem (40) verarbeitet werden, in dem nichtflüchtigen Speicher (37) inPartitionen protokolliert werden, wobei jede der Partitionen einer unterschiedlichen transaktionserzeugenden Dateneinheit zugeordnet ist.
     
    3. Verfahren nach Anspruch 1, wobei der Fixpunkt (52) in dem nichtflüchtigen Speicher (37) einen vollständigen Datensatz eines Heaps in dem flüchtigen Speicher (36) angibt, der von dem Transaktionsspeichersystem verwendet wird.
     
    4. Verfahren nach Anspruch 1, wobei der Vorgang des selektiven Speicherns das selektive Speichern der Daten in zumindest einen der folgenden Speicher umfasst:
    NAND-Flash-Speicher, Phasenwechselspeicher und/oder Memristorspeicher.
     
    5. Verfahren nach Anspruch 1, das ferner, als Reaktion auf einen Datenverlust in dem flüchtigen Speicher, ein Rekonstruieren Transaktionsspeicherdaten umfasst.
     
    6. Gegenstand, der wenigstens ein maschinenlesbares Speichermedium umfasst, das Anweisungen speichert, die bei Ausführung ein System mit wenigstens einem Prozessor zu Folgendem veranlassen:

    Verwenden eines flüchtigen Speichers (36) als Primärspeicher für Transaktionen, die einem Transaktionsspeichersystem zugeordnet sind; und

    selektives Speichern von Daten in einem nichtflüchtigen Speicher (37), um dem Transaktionsspeichersystem (40) Dauerhaftigkeit zu verleihen, um es dem Transaktionsspeichersystem (40) zu ermöglichen, im Falle eines Datenverlusts des flüchtigen Speichers in einen konsistenten Zustand zurückversetzt zu werden;

    wobei das maschinenlesbare Speichermedium Anweisungen speichert, die bei Ausführung das System mit dem wenigstens einen Prozessor ferner zu Folgendem veranlassen:

    Kopieren von in dem flüchtigen Speicher (36) gespeicherten Daten in einen ersten nichtflüchtigen Speicher(37)bereich, um einen Fixpunkt (52) zu erzeugen und Protokollieren von Datensätzen (54), die Transaktionen widerspiegeln, die nach der Erzeugung des Fixpunkts (52) in einem zweiten nichtflüchtigen Speicher(37)bereich aufgetreten sind;

    Zuordnen der gespeicherten Daten zu Ordnungskennungen, die von dem Transaktionsspeichersystem (40) Transaktionen zugewiesen werden, die die Daten erzeugt haben;

    Bereitstellen erster Bestätigungen an die Threads, die angeben, dass asynchrone Transaktionen in dem flüchtigen Speicher (36) gespeichert wurden, um das System in einem asynchronen Modus zu betreiben, und zweiter Bestätigungen an die Threads, die angeben, dass asynchrone Transaktionen in dem nichtflüchtigen Speicher (37) gespeichert wurden, um das System in einem synchronen Modus zu betreiben.


     
    7. Vorrichtung, die Folgendes umfasst:

    einen flüchtigen Speicher (36), um Transaktionsspeicherdaten zu speichern, die von mehreren Threads erstellt werden, die Transaktionen mit dem flüchtigen Speicher erzeugen;

    einen nichtflüchtigen Speicher; und

    ein Transaktionsspeichersystem (40), das für Folgendes angepasst ist:

    Verwenden des flüchtigen Speichers (36) als Primärspeicher für die Transaktionen; und

    selektives Speichern von Daten in einem nichtflüchtigen Speicher, um dem Transaktionsspeichersystem (40) Dauerhaftigkeit zu verleihen, um es dem Transaktionsspeichersystem (40) zu ermöglichen, im Falle eines Datenverlusts des flüchtigen Speichers in einen konsistenten Zustand zurückversetzt zu werden;

    wobei das Transaktionsspeichersystem (40) ferner zu Folgendem dient:

    Kopieren von in dem flüchtigen Speicher (36) gespeicherten Daten in einen ersten nichtflüchtigen Speicherbereich, um einen Fixpunkt (52) zu erzeugen und Protokollieren von Datensätzen (54), die Transaktionen widerspiegeln, die nach der Erzeugung des Fixpunkts (52) in einem zweiten nichtflüchtigen Speicherbereich aufgetreten sind;

    Zuordnen der gespeicherten Daten zu Ordnungskennungen, die von dem Transaktionsspeichersystem (40) Transaktionen zugewiesen werden, die die Daten erzeugt haben; und

    Bereitstellen erster Bestätigungen an die Threads, die angeben, dass asynchrone Transaktionen in dem flüchtigen Speicher (36) gespeichert wurden, um das System in einem asynchronen Modus zu betreiben, und zweiter Bestätigungen an die Threads, die angeben, dass asynchrone Transaktionen in dem nichtflüchtigen Speicher (37) gespeichert wurden, um das System in einem synchronen Modus zu betreiben.


     
    8. Vorrichtung nach Anspruch 7, wobei das Transaktionsspeichersystem (40) ferner angepasst ist, um die Protokolldatensätze (54) von Transaktionen zu protokollieren, die von dem Transaktionsspeichersystem (40) verarbeitet werden, in dem nichtflüchtigen Speicher (37) als Partitionen, wobei jede der Partitionen einer unterschiedlichen transaktionserzeugenden Dateneinheit zugeordnet ist.
     
    9. Vorrichtung nach Anspruch 7, wobei das Transaktionsspeichersystem (40) ferner angepasst ist, um den Fixpunkt (52) in dem nichtflüchtigen Speicher (37) zu speichern, der einen vollständigen Datensatz eines Heaps in dem flüchtigen Speicher (36) angibt, der von dem Transaktionsspeichersystem (40) verwendet wird.
     
    10. Vorrichtung nach Anspruch 7, wobei der nichtflüchtige Speicher (37) zumindest einen der folgenden Speicher umfasst:
    NAND-Flash-Speicher, Phasenwechselspeicher und/oder Memristorspeicher.
     
    11. Vorrichtung nach Anspruch 7, wobei das Transaktionsspeichersystem (40) ferner angepasst ist, um als Reaktion auf einen Datenverlust in dem flüchtigen Speicher Transaktionsspeicherdaten zu rekonstruieren.
     


    Revendications

    1. Procédé de mise en Ĺ“uvre d'un système de mémoire transactionnelle (40) comprenant :

    l'utilisation d'une mémoire volatile (36) comme stockage principal pour les transactions associées au système de mémoire transactionnelle (40) ; et

    le stockage sélectif de données dans une mémoire non volatile (37) pour conférer une durabilité au système de mémoire transactionnelle (40) afin de permettre au système de mémoire transactionnelle (40) d'être restauré en un état cohérent en cas de perte de données dans la mémoire volatile (36), dans lequel l'acte de stockage sélectif comprend :

    la copie des données stockées dans la mémoire volatile (36) dans une première région de mémoire non volatile (37) pour générer un point de contrôle (52) ;

    des enregistrements de journalisation (54) qui reflètent des transactions qui se sont produites après la génération du point de contrôle (52) dans une seconde région de mémoire non volatile (37) ;

    l'association des données stockées avec des identifiants de commande attribués par le système de mémoire transactionnelle (40) aux transactions qui ont généré les données ; et

    la fourniture d'une première indication à un fil que les données associées à une transaction ont été stockées dans la mémoire volatile (36) mais avant que les données aient été stockées dans la mémoire non volatile (37) alors que le système est en mode asynchrone et une seconde indication au fil que les données ont été stockées à la fois dans la mémoire volatile (36) et dans la mémoire non volatile (37) alors que le système est en mode synchrone.


     
    2. Procédé selon la revendication 1, dans lequel les enregistrements de journalisation (54) des transactions traitées par le système de mémoire transactionnelle (40) sont consignés dans la mémoire non volatile (37) en partitions, chacune des partitions étant associée à une entité génératrice de transactions différente.
     
    3. Procédé selon la revendication 1, dans lequel le point de contrôle (52) dans la mémoire non volatile (37) est indicatif d'un enregistrement complet d'un tas dans la mémoire volatile (36) utilisé par le système de mémoire transactionnelle.
     
    4. Procédé selon la revendication 1, dans lequel l'acte de stockage sélectif comprend le stockage sélectif des données dans au moins une des mémoires suivantes : une mémoire flash NAND, une mémoire à changement de phase et une mémoire memristor.
     
    5. Procédé selon la revendication 1, comprenant en outre, en réponse à une perte de données dans la mémoire volatile, la reconstruction des données de mémoire transactionnelle.
     
    6. Article comprenant au moins un support de mémoire exploitable sur machine stockant des instructions qui, lors de l'exécution, amènent un système ayant au moins un processeur à
    utiliser une mémoire volatile (36) comme stockage principal pour des transactions associées à un système de mémoire transactionnelle ; et stocker sélectivement des données dans une mémoire non volatile (37) pour conférer une durabilité au système de mémoire transactionnelle (40) afin de permettre au système de mémoire transactionnelle (40) d'être restauré en un état cohérent en cas de perte de données dans la mémoire volatile ; le support de mémoire exploitable sur machine stockant des instructions qui, lors de l'exécution, amènent le système ayant l'au moins un processeur à :

    copier des données stockées dans la mémoire volatile (36) dans une première région de mémoire non volatile (37) pour générer un point de contrôle (52) et des enregistrements de journalisation (54) qui reflètent des transactions qui se sont produites après la génération du point de contrôle (52) dans une seconde région de mémoire non volatile (37) ;

    associer les données stockées avec des identifiants de commande attribués par le système de mémoire transactionnelle (40) aux transactions qui ont généré les données ;

    fournir des premières confirmations aux fils indiquant que des transactions asynchrones ont été stockées dans la mémoire volatile (36) pour faire fonctionner le système en mode asynchrone et des secondes confirmations aux fils indiquant que des transactions asynchrones ont été stockées dans la mémoire non volatile (37) pour faire fonctionner le système en mode synchrone.


     
    7. Appareil comprenant : une mémoire volatile (36) pour stocker des données de mémoire transactionnelle créées par une pluralité de fils conducteurs générant des transactions avec la mémoire volatile ;
    une mémoire non volatile ; et
    un système de mémoire transactionnelle (40) conçu pour :

    utiliser la mémoire volatile (36) comme stockage principal pour les transactions ; et

    stocker sélectivement des données dans la mémoire non volatile pour conférer une durabilité au système de mémoire transactionnelle (40) afin de permettre au système de mémoire transactionnelle (40) d'être restauré en un état cohérent en cas de perte de données dans la mémoire volatile ; dans lequel

    le système de mémoire transactionnelle (40) est en outre :

    une copie des données stockées dans la mémoire volatile (36) dans une première région de mémoire non volatile pour générer un point de contrôle (52) et des enregistrements de journalisation (54) qui reflètent des transactions qui se sont produites après la génération du point de contrôle (52) dans une seconde région de mémoire non volatile ;

    une association des données stockées avec des identifiants de commande attribués par le système de mémoire transactionnelle (40) aux transactions qui ont généré les données ; et

    une fourniture des premières confirmations aux fils indiquant que des transactions asynchrones ont été stockées dans la mémoire volatile (36) pour faire fonctionner le système en mode asynchrone et des secondes confirmations aux fils indiquant que des transactions asynchrones ont été stockées dans la mémoire non volatile (37) pour faire fonctionner le système en mode synchrone.


     
    8. Appareil selon la revendication 7, dans lequel le système de mémoire transactionnelle (40) est en outre conçu pour enregistrer les enregistrements de journalisation (54) des transactions traitées par le système de mémoire transactionnelle (40) dans la mémoire non volatile (37) dans les partitions, chacune des partitions étant associée à une entité génératrice de transaction différente.
     
    9. Appareil selon la revendication 7, dans lequel le système de mémoire transactionnelle (40) est en outre conçu pour stocker le point de contrôle (52) dans la mémoire non volatile (37) indiquant un enregistrement complet d'un tas dans la mémoire volatile (36) utilisé par le système de mémoire transactionnelle (40).
     
    10. Appareil selon la revendication 7, dans lequel la mémoire non volatile (37) comprend au moins l'une des mémoires suivantes : une mémoire flash NAND, une mémoire à changement de phase et une mémoire memristor.
     
    11. Appareil selon la revendication 7, dans lequel le système de mémoire transactionnelle (40) est en outre conçu pour reconstruire des données de mémoire transactionnelle en réponse à une perte de données dans la mémoire volatile.
     




    Drawing




















    Cited references

    REFERENCES CITED IN THE DESCRIPTION



    This list of references cited by the applicant is for the reader's convenience only. It does not form part of the European patent document. Even though great care has been taken in compiling the references, errors or omissions cannot be excluded and the EPO disclaims all liability in this regard.

    Patent documents cited in the description