JP2017027326A - Storage system and program for the same - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a storage system capable of considerably reducing a data transfer amount from nodes (calculation devices), and a program for the storage system.SOLUTION: A storage system comprises: a plurality of nodes being calculation devices and individually provided with distributed databases; a transaction management device issuing a TID being an identifier of a transaction executed in each of the plurality of nodes, and holding information of a start of the transaction and information of an end of the transaction; and a nonvolatile memory type storage receiving an operation request of the transaction from each node, and notifying each node of operation completion after executing the transaction.SELECTED DRAWING: Figure 1

Description

  Embodiments described herein relate generally to a storage system and a storage system program.

  In a distributed database system in which multiple computing devices and storage devices are connected via a network, when multiple users request processing simultaneously from the storage device, there are many methods for improving the throughput processed per unit time. Transaction mechanisms using multi-version concurrency control (MVCC) are widely used. MVCC is a control method that resolves data access dependencies while maintaining multiple concurrency by maintaining multiple changes (versions) in an arbitrary record and ensuring consistency of information. .

  Conventionally, in the transaction mechanism using MVCC, in order to hold versions with multiple transaction IDs for the same record, even the same record is handled as different data, and it is necessary to acquire multiple versions in block units using the file system Naive records are being extracted. As described above, since data is transferred in units of blocks consisting of record groups, there is a problem that the load on the network is large.

JP 2011-204161 A

  The problem to be solved by the present invention is to provide a storage system and a storage system program capable of greatly reducing the amount of data transferred from a node (computing device).

  The storage system according to the embodiment issues a plurality of nodes each of which is a computing device and each having a distributed database, and a TID that is an identifier of a transaction executed on the node, and information on transaction start and transaction end And a non-volatile memory type storage that accepts a transaction operation request from the node and notifies the node of the completion of the operation after executing the transaction.

1 is a block diagram showing a schematic configuration of a storage system according to an embodiment of the present invention. It is a figure explaining a TID-logical address conversion list. It is a figure explaining the operation | movement at the time of writing of a new record. It is a flowchart which shows the flow of operation | movement at the time of writing of a new record. It is a figure explaining the operation | movement at the time of a record commit. It is a flowchart which shows the flow of operation | movement at the time of a record commit. It is a figure explaining the operation | movement at the time of the update of the existing record. It is a flowchart which shows the flow of operation | movement at the time of the update of the existing record. It is a figure explaining the operation | movement at the time of deletion of the existing record. It is a figure explaining the operation | movement at the time of the reading of the existing record. It is a flowchart which shows the flow of operation | movement at the time of the reading of the existing record. It is a figure explaining the operation | movement at the time of rollback limit TID registration. It is a figure explaining the operation | movement at the time of rollback. It is a figure explaining the operation | movement at the time of unnecessary TID deletion. It is a flowchart which shows the flow of operation | movement at the time of the update of the existing record.

  Hereinafter, an embodiment of the present invention will be described with reference to the drawings. In the drawings, the same portions are denoted by the same reference numerals, and redundant description is omitted.

  First, main terms used in the present embodiment will be described.

  “Database” refers to software that enables easy reuse, such as search and extraction, by writing and managing data in accordance with predetermined rules.

  “Storage” refers to a storage device that stores data.

  “File system” refers to software that manages data stored in a storage.

  “NAND flash memory” refers to a nonvolatile semiconductor memory device. In the present embodiment, it is preferable to configure the storage with a NAND flash memory.

  A “block” is a unit for reading / writing data of a file system.

  “Record” means a unit of data registration in a database.

  “Record ID” refers to an identifier that is sequentially assigned to each record.

  “Transaction processing” refers to processing that handles a plurality of data operations as an inseparable series of operations. Transaction processing includes elements of transaction ID (TID), commit, and rollback.

  The “transaction ID (TID)” is an identifier that is assigned sequentially for each transaction process. A larger TID indicates a new transaction, and a smaller TID indicates an old transaction.

  “Commit” is to confirm a series of operations related to transaction processing.

  “Rollback” refers to rewinding the processing in units of transaction processing to the previous state.

  The “transaction management device” is a device that issues TIDs of transactions executed at each node and holds information on the start and end of transactions.

  “Snapshot” refers to a set of TIDs that have already been committed at the start of a transaction. It is managed by the transaction management device and is passed to the database each time.

  In this embodiment, in a distributed database having MVCC, a non-volatile memory storage determines and transfers a TID record suitable for a transaction to be accessed without using a file system.

  FIG. 1 is a block diagram showing a schematic configuration of a storage system according to an embodiment of the present invention. This apparatus is realized using a general-purpose computer (for example, a personal computer (PC) or the like) and software operating on the computer. The computer includes an engineering workstation (EWS) suitable for CAD (Computer Aided Design) and CAE (Computer Aided Engineering). In this computer, in this embodiment, a TID that is an identifier of a transaction executed in a node is added to a storage system that is a computing device and each includes a plurality of nodes each including a distributed database and a nonvolatile memory type storage. To store the transaction start information and transaction end information, and to receive the transaction operation request from the node, execute the transaction, and notify the node of the operation completion. It can also be implemented as a storage system program.

  As shown in FIG. 1, the storage system 1 according to the present embodiment mainly includes a nonvolatile memory storage 10, a transaction management device 20, and a plurality of nodes 30.

  The nonvolatile memory type storage 10 receives an operation request such as record writing from the node 30 and executes the processing. Details of the configuration of the nonvolatile memory type storage 10 and the processing for each operation request from the node 30 will be described later.

  The transaction management device 20 issues a TID of a transaction executed on each node 30 and holds information on the start of the transaction and information on the end of the transaction.

  The node 30 is a computing device and includes a distributed database. A distributed database has a hardware configuration in which a plurality of computing devices (nodes) and storage devices are connected via a network, and is a system in which all nodes can access records in a shared storage device via a network.

  The nonvolatile memory type storage 10 includes a transaction record access unit 11, a rollback limit TID holding unit 12, a node access information holding unit 13, a TID-logical address conversion list 14, a logical-physical conversion table 15, and a nonvolatile memory 16. Yes.

  The transaction record access unit 11 provides the following functions to each node 30. The functions (5) to (7) are characteristic functions in this embodiment that do not use a conventional file system.

(1) Record write operation: The record ID and the TID being executed are used as arguments.
(2) Record commit operation: The record ID and the TID being executed are used as arguments.
(3) Record read operation: The record ID and the TID being executed are used as arguments.
(4) Record deletion operation: The record ID and the TID being executed are used as arguments.
(5) Rollback operation: Cancels the commit operation and returns to the previous rewind state in units of transactions. The argument is the TID to which to rewind to the previous state.
(6) Rollback limit TID registration operation: TID that is the rewind limit is used as an argument.
(7) Unnecessary TID deletion operation: An unnecessary TID is deleted without an argument.

  When receiving these operation requests from the node 30, the transaction record access unit 11 sends the request source node access information to the node access information holding unit 13. When the request is completed, the transaction record access unit 11 transmits a completion notification using the request source node access information held in the node access information holding unit 13.

  The rollback limit TID holding unit 12 holds a TID that is a limit for rewinding transaction processing.

  The node access information holding unit 13 holds a node number, a record ID, and a TID for the operation request source node for the nonvolatile memory storage 10.

  The TID-logical address conversion list 14 holds the relationship between the record ID and the TID via the logical address in order to hold the relationship between the written data and TID necessary for data extraction. For each ID, each element of “operated TID”, “logical address”, “commit flag”, and “delete flag” is stored in a list format. FIG. 2 is a diagram for explaining the TID-logical address conversion list 14. Here, the operated TID is a TID obtained by adding an element to the list. The logical address is the logical address where the operated TID has written the record. The commit flag indicates whether or not the commit has been completed. There are “True” indicating a committed state and “False” indicating a committed state. The deletion flag indicates whether data has been deleted. There are “True” indicating a deleted state and “False” indicating a state where deletion has not been completed.

  The logical-physical conversion table 15 stores a correspondence relationship between a physical address and a logical address in a storage area of the nonvolatile memory 16 that stores data.

  As the non-volatile memory 16, a NAND flash memory which is easy to increase in capacity at a high speed is suitable. A NAND flash memory is a nonvolatile semiconductor memory that can electrically rewrite data and can retain data without supplying power. The nonvolatile memory 16 has a unique physical address.

  Next, processing for each operation request in the storage system 1 configured as described above will be described in detail.

<Write new record>
FIG. 3 is a diagram for explaining the operation when writing a new record. In FIG. 3, the operation will be described by taking as an example a case where a write request of record ID = 1 and TID = 100 is received from the node A. The transaction record access unit 11 that has received the request to write a new record registers node A, record ID = 1, and TID = 100 as node access information in the node access information holding unit 13.

  Subsequently, the transaction record access unit 11 registers the TID-logical address in the TID-logical address conversion list 14. In the example shown in FIG. 2, the operated TID = 100 and logical address = 1 are registered. In this state, since the transaction is not confirmed, the commit flag = F is registered, and the deletion flag = F is registered because it is a write request. .

  Subsequently, the transaction record access unit 11 registers the physical address of the nonvolatile memory 16 in the logical address = 1 of the logical-physical conversion table 15.

  Subsequently, the transaction record access unit 11 writes a record at a predetermined physical address in the nonvolatile memory 16.

  The transaction record access unit 11 refers to the node access information registered in the node access information holding unit 13 and transmits a write completion notification to the node A.

  FIG. 4 is a flowchart showing the flow of operations when writing a new record.

  First, the transaction record access unit 11 of the nonvolatile memory storage 10 receives a request for writing a new record (step S41).

  Next, the node number, record ID, and TID requested to be written are registered in the node access information holding unit 13 (step S42).

  Next, the TID-logical address conversion list 14 is searched for the presence or absence of a list that matches the record ID (step S43).

  If the list does not exist, a new list is created for the record ID (step S44), and the process proceeds to step S45. Since the list of target record IDs does not exist when writing a new record, the process goes through the flow of step S44.

  On the other hand, if the list exists, elements are added to the list so that the TIDs are in ascending order, the TID and logical address are registered, and the commit flag and the deletion flag are registered as False (step S45).

  Subsequently, the physical address is registered in the logical address of the logical-physical conversion table 15 (step S46).

  Next, a record is written at a predetermined physical address in the nonvolatile memory 16 (step S47).

  Subsequently, the requested node is retrieved from the node access information holding unit 13 based on the record ID and TID (step S48).

  Next, the transaction record access unit 11 transmits a write completion notification to the node (step S49), and ends the process at the time of writing a new record.

<Record commit>
FIG. 5 is a diagram for explaining the operation at the time of record commit. In FIG. 5, the operation will be described by taking as an example a case where a commit request of record ID = 1 and TID = 100 is received from the node A. The transaction record access unit 11 that has received the commit request registers node A, record ID = 1, and TID = 100 as node access information in the node access information holding unit 13.

  Subsequently, the transaction record access unit 11 registers the TID-logical address in the TID-logical address conversion list 14. In the example shown in FIG. 5, a change to commit flag = T is registered.

  Next, the transaction record access unit 11 refers to the node access information registered in the node access information holding unit 13 and transmits a record commit completion notification to the node A.

  FIG. 6 is a flowchart showing the flow of operations at the time of record commit.

  First, the transaction record access unit 11 of the nonvolatile memory type storage 10 receives a request to commit a record (step S61).

  Next, the node number, record ID, and TID for which a commit request is made are registered in the node access information holding unit 13 (step S62).

  Subsequently, the record ID and TID that match are searched from the TID-logical address conversion list 14, and the commit flag is changed to True (step S63).

  Next, based on the record ID and TID, the node that made the commit request is searched from the node access information holding unit 13 (step S64).

  Next, the transaction record access unit 11 transmits a record commit completion notification to the node (step S65), and ends the process at the time of record commit.

<Updating existing records>
FIG. 7 is a diagram for explaining the operation when updating an existing record. In FIG. 7, an operation when an update request of record ID = 1 and TID = 103 is received from the node A will be described as an example. Upon receiving the update write request, the transaction record access unit 11 registers node A, record ID = 1, and TID = 103 as node access information in the node access information holding unit 13.

  Next, the transaction record access unit 11 registers the TID-logical address in the TID-logical address conversion list 14 by sorting in ascending order in the TID order. In the example shown in FIG. 7, the operated TID = 103 and logical address = 2 are registered. In this state, since the transaction is not confirmed, the commit flag = F is registered, and the deletion flag = F is registered because it is an update write request. Is done.

  Subsequently, the transaction record access unit 11 registers the physical address of the nonvolatile memory 16 in the logical address = 2 of the logical / physical conversion table 15.

  Subsequently, the transaction record access unit 11 writes a record at a predetermined physical address in the nonvolatile memory 16.

  Next, the transaction record access unit 11 refers to the node access information registered in the node access information holding unit 13 and transmits an update write completion notification to the node A.

  FIG. 8 is a flowchart showing the flow of operations when updating an existing record.

  First, the transaction record access unit 11 of the nonvolatile memory storage 10 receives a request for updating an existing record (step S81).

  Next, the node number, record ID, and TID for which an existing record update request has been made are registered in the node access information holding unit 13 (step S82).

  Next, a list whose record ID matches the TID-logical address conversion list 14 is searched (step S83).

  If the list does not exist, a new list is created for the record ID (step S84), and the process proceeds to step S85. When the existing record is updated, the list of target record IDs exists, so that the TID is in ascending order. An element is added to the list, a TID and a logical address are registered, and a commit flag and a deletion flag are registered as False (step S85).

  Next, the physical address of the nonvolatile memory 16 is registered in the logical address of the logical-physical conversion table 15 (step S86).

  Next, a record is written at a predetermined physical address in the nonvolatile memory 16 (step S87).

  Subsequently, the node that requested the update of the existing record is searched from the node access information holding unit 13 based on the record ID and TID (step S88).

  Next, the transaction record access unit 11 transmits an update completion notification of the existing record to the node (step S89), and ends the process at the time of updating the existing record.

<Delete existing record>
FIG. 9 is a diagram for explaining the operation when deleting an existing record. In FIG. 9, the operation when receiving a deletion request of record ID = 1 and TID = 108 from the node A will be described as an example. The transaction record access unit 11 that has received a request to delete an existing record registers node A, record ID = 1, and TID = 108 as node access information in the node access information holding unit 13.

  Subsequently, the transaction record access unit 11 registers the TID-logical address in the TID-logical address conversion list 14 by sorting in ascending order in the TID order. In the example shown in FIG. 9, the operated TID = 108 is registered. In this state, since the transaction is not confirmed, the commit flag = F is registered, and since it is a deletion request, True is registered in the deletion flag.

  The transaction record access unit 11 refers to the node access information registered in the node access information holding unit 13 and transmits a deletion completion notification of the existing record to the node A.

  Note that the same commit operation as that at the time of record commit is applied to the completion of the delete operation of the existing record.

<Read existing record>
FIG. 10 is a diagram for explaining the operation when reading an existing record. In FIG. 10, the operation when a read request of record ID = 1 and TID = 105 is received from the node B will be described as an example. The transaction record access unit 11 that has received the read request for the existing record registers the node B, the record ID = 1, and the TID = 105 as the node access information in the node access information holding unit 13.

  Subsequently, the transaction record access unit 11 searches for the TID-logical address in the TID-logical address conversion list 14, but searches for the latest TID below the TID of the read request. In the example shown in FIG. 10, the operated TID = 103, logical address = 2, commit flag = T, and deletion flag = F are extracted.

  Therefore, the transaction record access unit 11 acquires the physical address of logical address = 2 in the logical / physical conversion table 15.

  Subsequently, the transaction record access unit 11 reads a record from a predetermined physical address in the nonvolatile memory 16.

  The transaction record access unit 11 refers to the node access information registered in the node access information holding unit 13 and transmits an existing record read completion notification to the node B.

  FIG. 11 is a flowchart showing the flow of operations when reading an existing record.

  First, the transaction record access unit 11 of the nonvolatile memory storage 10 receives a request to read an existing record (step S1101).

  Next, the node number, the record ID, and the TID for which the request for reading the existing record is requested are registered in the node access information holding unit 13 (step S1102).

  Next, the presence / absence of a list whose record ID matches the TID-logical address conversion list 14 is searched (step S1103).

  If the list does not exist (step S1104), an error occurs.

  On the other hand, if the list exists, the element of the maximum operation TID that is not more than the operation TID read from the list and is TID is searched (step S1105).

  Next, it is determined whether the commit flag is true and the delete flag is false (step S1106).

  If No in step S1106, the element of the one older TID is read (step S1107).

  Next, it is determined whether the commit flag is true and the delete flag is false (step S1108).

  If No in step S1108, the process returns to step S1107. If Yes in step S1108, the process proceeds to step S1109.

  If Yes in step S1106, the logical address is read from the list, and the physical address of the nonvolatile memory 16 is read from the corresponding logical address in the logical-physical conversion table (step S1109).

  Next, a record is read from a predetermined physical address in the nonvolatile memory 16 (step S1110).

  Subsequently, the requested node is searched from the node access information holding unit 13 based on the record ID and TID (step S1111).

  Next, the transaction record access unit 11 transmits a read completion notification and the read record to the node (step S1112), and ends the process when reading the existing record.

<Rollback limit TID registration>
FIG. 12 is a diagram for explaining the operation when registering the rollback limit TID. In FIG. 12, the operation when a registration request for the rollback limit TID is received from the node A will be described as an example.

  Upon receiving the rollback limit TID registration request, the transaction record access unit 11 registers the node A and the rollback limit TID as node access information in the node access information holding unit 13.

  Subsequently, the transaction record access unit 11 registers the rollback limit TID in the rollback limit TID holding unit 12.

<Rollback>
Next, a description will be given of a rollback for making a previous rewind state in a transaction unit. FIG. 13 is a diagram for explaining the operation during rollback. In FIG. 13, the operation when receiving a rollback request of TID = 105 from the node B will be described as an example.

  Upon receiving the rollback request, the transaction record access unit 11 registers the node B, TID = 105, and rollback as node access information in the node access information holding unit 13.

  Subsequently, the transaction record access unit 11 accesses the rollback limit TID holding unit 12 and compares the TID that requested the rollback with the registered rollback limit TID.

  Next, the transaction record access unit 11 searches the TID-logical address conversion list 14 for a list in which the TID requested for the rollback matches.

  Since the rewind state has been entered, in the example shown in FIG. 13, when the operated TID = 108, the commit flag is changed to F.

<Delete unnecessary TID>
FIG. 14 is a diagram for explaining the operation when deleting unnecessary TIDs. In FIG. 14, an operation when an unnecessary TID deletion request is received from the node A will be described as an example. The transaction record access unit 11 that has received the unnecessary TID deletion request registers node A and unnecessary TID deletion as node access information in the node access information holding unit 13.

  Next, the transaction record access unit 11 reads the rollback limit TID from the rollback limit TID holding unit 12.

  Subsequently, the transaction record access unit 11 selects unnecessary data from the TID-logical address conversion list 14.

  Next, the transaction record access unit 11 acquires a physical address of the nonvolatile memory 16 corresponding to unnecessary data in the logical-physical conversion table 15.

  Subsequently, the transaction record access unit 11 deletes the record from the predetermined physical address of the nonvolatile memory 16.

  The transaction record access unit 11 refers to the node access information registered in the node access information holding unit 13 and transmits an unnecessary TID deletion completion notification to the node A.

  FIG. 15 is a flowchart showing the flow of operations when updating an existing record.

  First, the transaction record access unit 11 of the nonvolatile memory storage 10 receives a request for deleting unnecessary TIDs (step S1501).

  Next, the node number for which unnecessary TID deletion was requested and the execution of unnecessary TID deletion are registered in the node access information holding unit 13 (step S1502).

  Next, the rollback limit TID is read (step S1503).

  Next, a list of record IDs registered first is read from the TID-logical address conversion list 14 (step S1504).

  Subsequently, an element is read from the list (step S1505).

  Next, from the rollback limit TID among the elements of the list, one old TID is deleted from the TID less than or equal to the rollback limit TID (step S1506).

  Next, the data of the physical address of the nonvolatile memory 16 pointed to by the logical address included in the deleted TID is deleted from the memory (step S1507).

  Subsequently, it is determined whether or not the next record ID exists in the list (step S1508).

  If the next record ID exists in the list (Yes in step S1508), the list of the next record ID is read (step S1509), and the process returns to step S1505.

  On the other hand, if the next record ID does not exist in the list (No in step S1508), the node that requested the unnecessary TID deletion is searched from the node access information holding unit 13 (step S1510).

  Next, the transaction record access unit 11 transmits a notification of completion of unnecessary TID deletion to the node (step S1511), and ends the processing at the time of unnecessary TID deletion.

  As described above, according to the present embodiment, the amount of data transferred from the computing device can be greatly reduced, the throughput can be improved, and the load on the network can be reduced. For example, assuming that each record has an average of 10 versions, the data transfer amount from each node to the storage can be reduced to 1/10, so that the throughput is improved 10 times.

  Although several embodiments of the present invention have been described, these embodiments are presented by way of example and are not intended to limit the scope of the invention. These novel embodiments can be implemented in various other forms, and various omissions, replacements, and changes can be made without departing from the scope of the invention. These embodiments and modifications thereof are included in the scope and gist of the invention, and are included in the invention described in the claims and the equivalents thereof.

DESCRIPTION OF SYMBOLS 1 ... Storage system 10 ... Non-volatile memory type storage 11 ... Transaction record access part 12 ... Rollback limit TID holding part 13 ... Node access information holding part 14 ... TID-logical address Conversion list 15 ... logical-physical conversion table 16 ... non-volatile memory 20 ... transaction management device 30 ... node

Claims (7)

A plurality of nodes each of which is a computing device and each has a distributed database;
A transaction management device that issues a TID, which is an identifier of a transaction executed in the node, and holds transaction start information and transaction end information;
A non-volatile memory type storage that accepts a transaction operation request from the node and notifies the node of the completion of the operation after executing the transaction;
A storage system comprising: The nonvolatile memory storage is
For the node, a record write operation, a record commit operation for confirming an operation related to a transaction, a record read operation, a record delete operation, a rollback operation for setting a previous rewind state in a transaction unit, and a rewind limit Transaction record access part that provides functions for rollback limit TID registration operation and unnecessary TID deletion operation,
A rollback limit TID holding unit for holding the TID which is a limit of rewinding a transaction;
Regarding the operation request source node for the nonvolatile memory type storage, a node number, a record ID that is an identifier of the record, a node access information holding unit that holds the TID,
A TID-logical address conversion list that holds the relationship between the record ID and the TID via a logical address;
A non-volatile memory for storing the record;
A logical-physical conversion table storing a correspondence relationship between a physical address of the storage area of the nonvolatile memory and the logical address;
The storage system according to claim 1.   When the transaction record access unit accepts an operation request from the node, the operation request source node access information is sent to the node access information holding unit, and when the request is completed, the node access information holding unit holds the node access information. The storage system according to claim 2, wherein a completion notification is transmitted to the node using the node access information of the operation request source.   The storage system according to any one of claims 1 to 3, wherein the nonvolatile memory is a NAND flash memory having a unique physical address. The TID-logical address conversion list is
The operated TID representing the TID with the element added to the list, the logical address at which the operated TID has written the record, the commit flag indicating whether or not the commit has been completed, and whether or not the record has been deleted The storage system according to any one of claims 2 to 4, wherein each element of the deletion flag is held in a list format. The rollback operation takes as an argument the TID to which to rewind to the previous state,
The rollback limit TID registration operation takes the TID as the rewind limit as an argument,
The storage system according to any one of claims 2 to 5, wherein the unnecessary TID deletion operation does not involve an argument. A storage system comprising a plurality of nodes each having a distributed database and a non-volatile memory type storage.
A function for issuing a TID that is an identifier of a transaction executed in the node, and holding information on the start of a transaction and information on the end of a transaction;
A function of accepting a transaction operation request from the node and executing the transaction and notifying the node of the completion of the operation;
Storage system program for realizing

Images