Distributed multi-version commitment ordering protocols for guaranteeing serializability during transaction processing

• US 5,701,480 A
• Filed: 04/14/1993
• Issued: 12/23/1997
• Est. Priority Date: 10/17/1991
• Status: Expired due to Term

• Current Assignee: Hewlett-Packard Development Company, L.P. (HP Inc.)
• Original Assignee: Digital Equipment Corporation (HP Inc.)
• Inventors: Raz, Yoav
• Primary Examiner(s): Kriess, Kevin A.
• Assistant Examiner(s): Chaki, Kakali
• Application Number:
  US08/047,271
• Time in Patent Office: 1,714 Days
• Field of Search: 395/700, 395/650, 395/600, 395/610, 395/671, 395/182.17
• US Class Current: 718/101
• CPC Class Codes: G06F 9/466 Transaction processing

Abstract

In a multi-version database, copies of prior committed versions (snapshots) are kept for access by the read-only transactions. The read-write transactions are selectively aborted to enforce an order of commitment of read-write transactions that is the same as an order of conflicts among the read-write transactions. In a preferred embodiment, the read-write transactions are serialized by maintaining and referencing a graph of conflicts among read-write transactions, and the read-only transactions are serialized by a timestamp mechanism for selection of the snapshots to be read. Each time that a read-write transaction is committed, the read-write transaction is assigned a unique timestamp that is used to timestamp all resources committed by the read-write transaction. Upon starting, each read-only transaction is also assigned a timestamp. Each read-only transaction reads only the latest committed versions of all resources, that are timestamped earlier than the timestamp of the read-only transaction. In a multiprocessing system, the timestamps are issued to global coordinators and distributed locally with atomic commit messages and global queries. Moreover, read-write transactions may selectively access a hierarchy of uncommitted versions to prepare for various possible commitment orders. The hierarchy defines a path for record access and for cascading aborts. A plurality of mutually-conflicting uncommitted versions may be prepared for each transaction to prepare for all possible commitment orders.

Claims

1. A method of operating a digital computer to process read-write transactions and read-only transactions in a computer system, said method comprising the steps of:

a) beginning preparation of results of said transactions;

b) determining an order of conflicts among said read-write transactions;

c) committing to memory state of said computer system prepared results of a selected one of said read-write transactions;

d) aborting an abort set of said read-write transactions for which commitment is contrary to said order of conflicts and said committing to memory state of said computer system said prepared results of said selected one of said read-write transactions;

e) retaining a prior version of memory state of said computer system existing prior to being updated by said prepared results of said selected one of said read-write transactions; and

f) permitting selected ones of said read-only transactions to read said prior version of memory state after said prepared results of said selected one of said read-write transactions are committed to memory state of said computer system, while preventing said read-write transactions from reading said prior version of memory state after said prepared results'"'"' of said selected one of said read-write transactions are committed to memory state of said computer system.

2. The method as claimed in claim 1, wherein said step of determining said order of conflicts includes detecting when a data access operation for one of said read-write transactions addresses data accessed by data access operations for other ones of said read-write transactions.

3. The method as claimed in claim 1, wherein said read-only transactions are prevented from reading any results of said read-write transactions before said results of said read-write transactions are committed to said state memory, so that said read-only transactions do not conflict with said read-write transactions.

4. The method as claimed in claim 1, wherein said read-write transactions include global read-write transactions that are distributed across the computing system and local read-write transactions that are not distributed across the computing system, and said step of determining an order of conflicts among said read-write transactions disregards conflicts between said local read-write transactions but takes into account indirect conflicts between said global read-write transactions that are caused by said local read-write transactions.

5. The method as claimed in claim 1, wherein said step of determining an order of conflicts comprises recording in memory of said digital computer a graph of conflict orders between said transactions, and wherein said method further comprises searching said graph for determining said abort set of transactions, said abort set including a transaction on each path in said graph to said selected one of said global transactions from a global transaction not yet having results aborted or committed to memory state of said computer system.

6. The method as claimed in claim 1, further comprising the step of selecting said selected one of said read-write transactions in response to a commit command from a global coordinator.

7. The method as claimed in claim 1, further comprising the step of selecting said selected one of said read-write transactions in order to minimize the number of read-write transactions that are aborted in said step d).

8. The method as claimed in claim 1, further comprising the step of receiving from a coordinator a request to prepare a specified one of said read-write transactions, and delaying acknowledgement of completion of preparation of said specified one of said read-write transactions until none of said transactions not yet committed in step c) nor aborted in step d) are contrary to said order of conflicts and commitment of said specified one of said read-write transactions.

9. The method as claimed in claim 1, wherein a read operation of a second one of said read-write transactions reads write data written by a write operation of a first one of said read-write transactions before said first one of said read-write transactions is committed, and wherein said method further comprises the step of aborting results of all of said transactions that have read data written by aborted read-write transactions.

10. The method as claimed in claim 1, wherein a sufficient number of prior committed versions of memory state of said computer system are retained to process said read-only transactions by permitting each read-only transaction to read a version of memory state last committed prior to the time that processing of said each read-only transaction is begun and without delaying commitment of read-write transactions that update memory state to be read by said each read-only transaction.

11. The method as claimed in claim 1, wherein said computer system includes a plurality of processors, said read-write transactions include global read-write transactions that are distributed across the computer system and that are each processed and committed at more than one of said processors, and said read-only transactions include global read-only transactions that are distributed across the computer system and are each processed at more than one of said processors, and said method further includes synchronizing the processing of said global read-only transactions with commitment of results of said global read-write transactions at each of said processors so that each global read-only transaction reads results committed at one of said processors that are consistent with results committed, and read by said each global read-only transaction, at another of said processors.

12. The method as claimed in claim 11, wherein said synchronizing is performed by synchronizing transmission of commitment messages from a global update coordinator to said processors with initiation of said each global read-only transaction by a global query coordinator.

13. The method as claimed in claim 11, wherein said synchronizing is performed by serializing performance of said global read-only transactions with commitment of said read-write transactions.

14. The method as claimed in claim 11, wherein the performance of said global read-only transactions is serialized with the commitment of said read-write transactions by issuing a timestamp when each of said global read-write transactions is committed and when each of said global read-only transactions is initiated, and for each global read-only transaction, comparing the timestamp issued to said each global read-only transaction with timestamps issued to global read-write transactions to determine a proper version of said memory state of said computer system to be read by said each global read-only transaction.

15. A method of operating a digital computer to process read-write transactions and read-only transactions in a computer system, said method comprising the steps of:

a) beginning preparation of results of said transactions;

b) determining an order of conflicts among said read-write transactions by detecting when a data access operation for one of said read-write transactions addresses data accessed by data access operations for other ones of said read-write transactions;

c) committing to memory state of said computer system prepared results of a selected one of said read-write transactions;

d) aborting an abort set of said read-write transactions for which commitment is contrary to said order of conflicts and said committing to memory state of said computer system said prepared results of said selected one of said read-write transactions;

e) retaining a prior version of memory state of said computer system existing prior to being updated by said prepared results of said selected one of said read-write transactions; and

f) maintaining a sufficient number of prior committed versions of memory state of said computer system to process said read-only transactions by permitting each read-only transaction to read a version of memory state last committed prior to the time that processing of said each read-only transaction is begun and without delaying commitment of read-write transactions that update memory state to be read by said each read-only transaction, while preventing said read-write transactions from reading any of said prior versions of memory state having been updated by committed ones of said read-write transactions, and preventing said read-only transactions from reading any results of said read-write transactions before said results of said read-write transactions are committed to said memory state of said computer system, so that said read-only transactions do not conflict with said read-write transactions.

16. The method as claimed in claim 15, wherein said computer system includes a plurality of processors, said read-write transactions include global read-write transactions that are distributed across the computing system and that are each processed and committed at more than one of said processors, and said read-only transactions include global read-only transactions that are distributed across the computing system and that are each processed at more than one of said processors, and said method further includes synchronizing the processing of said global read-only transactions with commitment of results of said global read-write transactions at each of said processors so that each global read-only transaction reads results committed at one of said processors that are consistent with results committed, and read by said each global read-only transaction, at another of said processors.

17. The method as claimed in claim 16, wherein said synchronizing is performed by synchronizing transmission of commitment messages from a global update coordinator to said processors with initiation of said each global read-only transaction by a global query coordinator.

18. The method as claimed in claim 16, wherein performance of said global read-only transactions is synchronized and serialized with commitment of said read-write transactions by issuing a timestamp when each of said global read-write transactions is committed and when each of said global read-only transactions is initiated, and for each global read-only transaction, comparing the timestamp issued to said each global read-only transaction with timestamps issued to global read-write transactions to determine the version of memory state last committed prior to the time that processing of said each global read-write transaction is begun.

19. A digital computer system for processing read-write transactions and read-only transactions, said digital computer system comprising, in combination:

a) means for performing operations of said read-write transactions such that operations of some read-write transactions are performed in accordance with availability of resources of said digital computer system before commitment of other read-write transactions;

b) means for determining an order of conflicts among said read-write transactions; and

c) means for enforcing an order of commitment of selected ones of said read-write transactions in accordance with said order of conflicts, said means for enforcing including means for delaying commitment of selected read-write transactions and means for aborting an abort set of said read-write transactions selected so that committing of a selected one of said read-write transactions before commitment of other of said read-write transactions excluded from said abort set is consistent with said order of conflicts; and

d) means for maintaining a sufficient number of prior committed versions of memory state of said computer system to process said read-only transactions by permitting each read-only transaction to read a version of memory state last committed prior to the time that processing of said each read-only transaction is begun and without delaying commitment of read-write transactions that update memory state to be read by said each read-only transaction, while preventing said read-write transactions from reading any prior versions of memory state having been updated by committed ones of said read-write transactions, and preventing said read-only transactions from reading any results of said read-write transactions before said results of said read-write transactions are committed to said memory state, so that said read-only transactions do not conflict with said read-write transactions.

20. The digital computer system as claimed in claim 19, further including means for delaying the aborting of read-write transactions in said abort set when said means for delaying commitment delays commitment of said selected one of said read-write transactions.

21. The digital computer system as claimed in claim 19, wherein said means for delaying includes means for delaying acknowledgement of completion of preparation of a selected one of said read-write transactions until committing of said specified one of said read-write transactions before committing all other of said read-write transactions not yet committed nor aborted is consistent with said order of conflicts.

22. The digital computer system as claimed in claim 21, further comprising means for terminating said delaying in response to a signal in an atomic commitment protocol.

23. The digital computer system as claimed in claim 19, wherein said means for aborting includes means for aborting all of said read-write transactions that have read data written by aborted read-write transactions.

24. The digital computer system as claimed in claim 19, wherein said means for determining said order of conflicts comprises means for recording in memory of said digital computer system a graph of conflict orders between said read-write transactions, and means for searching said graph for determining said abort set of read-write transactions, said abort set including a read-write transaction on each path in said graph to said selected one of said read-write transactions from a read-write transaction not yet having results aborted or committed to memory state of said digital computer system.

25. The digital computer system as claimed in claim 19, wherein said read-write transactions include global read-write transactions that are distributed across said computer system and that are each committed at more than one of said processors, and local read-write transactions that are not distributed across said computer system and that are each committed at only one of said processors, and wherein said digital computer system includes means employing write-locks for ensuring that local read-write transactions provide consistent results but permitting global transactions to disregard said write-locks, and wherein said means for determining an order of conflicts disregards conflicts between said local read-write transactions but takes into account indirect conflicts between said global read-write transactions caused by said local read-write transactions.

26. The digital computer system as claimed in claim 19, wherein said digital computer system includes a plurality of processors, said read-write transactions include global read-write transactions that are distributed across said processors and that are each processed and committed at more than one of said processors, and said read-only transactions include global read-only transactions that are distributed across said processors and that are each processed at more than one of said processors, and said computer system further includes means for synchronizing the processing of said global read-only transactions with commitment of results of said global read-write transactions at each of said processors so that each global read-only transaction reads results committed at one of said processors that are consistent with results committed, and read by said each global read-only transaction, at another of said processors.

27. The digital computer system as claimed in claim 26, wherein said means for synchronizing includes means for synchronizing transmission of commitment messages from a global update coordinator to said processors with initiation of said each global read-only transaction by a global query coordinator.

28. The digital computer system as claimed in claim 26, wherein said means for synchronizing includes means for serializing processing of said global read-only transactions with commitment of said read-write transactions by issuing a timestamp when each of said global read-write transactions is committed and when each of said global read-only transactions is initiated, and for each global read-only transaction, comparing the timestamp issued to said each global read-only transaction with timestamps issued to global read-write transactions to determine the version of memory state last committed prior to the time that processing of said each global read-write transaction is begun.

29. A method of operating a digital computer to process read-write transactions in a computer system, said method comprising the steps of:

a) beginning preparation of results of said read-write transactions, said results including uncommitted versions of memory state associated with said read-write transactions, and

(i) preparing a first transaction by modifying a last committed version of memory state to create a first uncommitted version of memory state associated with said first transaction, and

(ii) before committing said first uncommitted version to memory state of said computer system, preparing a second transaction by modifying said first uncommitted version of memory state to create a second uncommitted version of memory state associated with said second transaction;

b) determining an order of conflicts among said versions of memory state associated with said read-write transactions, wherein said order of conflicts includes said second uncommitted version of memory state following said first uncommitted version of memory state;

c) committing to memory state of said computer system prepared results of a selected one of said read-write transactions; and

d) aborting an abort set of said uncommitted versions of memory state for which commitment is contrary to said order of conflicts and said committing to memory state of said computer system said prepared results of said selected one of said read-write transactions, and when said first uncommitted version of memory state is aborted, aborting said second uncommitted version.

30. The method as claimed in claim 29, wherein said beginning preparation of results of said read-write transactions includes following a predefined strategy for selecting either said last committed version or one of said uncommitted versions for access by a third transaction for which preparation has begun.

31. The method as claimed in claim 30, wherein said predefined strategy includes initially selecting said last committed version, and from said last committed version preparing a third uncommitted version associated with said third transaction, and after a conflict is detected between said third uncommitted version and said first uncommitted version wherein said third uncommitted version is based upon a resource of said last committed version having been modified in said first uncommitted version, preparing from said first uncommitted version a fourth uncommitted version associated with said third transaction.

32. A method of operating a digital computer to process read-write transactions in a computer system, said method comprising the steps of:

a) beginning preparation of results of said read-write transactions, said results including uncommitted versions of memory state associated with said transactions, including;

(i) preparing a first transaction by modifying a last committed version of memory state to create a first uncommitted version of memory state associated with said first transaction, and

(ii) preparing a second transaction by modifying said last committed version of memory state to create a second uncommitted version of memory state associated with said second transaction;

b) during preparation of said uncommitted versions, detecting conflicts between said uncommitted versions, said conflicts occurring when one uncommitted version includes a modified version of a resource of a prior version and another uncommitted version is based upon said resource of said prior version, wherein said first uncommitted version includes a modified version of a resource in said last committed version upon which said second uncommitted version is based, and said second uncommitted version includes a modified version of a resource in said last committed version upon which said first uncommitted version is based, so that said first uncommitted version and said second uncommitted version are mutually conflicting;

c) committing to memory state of said computer system an uncommitted version of memory state associated with a selected one of said read-write transactions; and

d) aborting an abort set of said uncommitted versions of memory state which are based upon any resources of any prior version having been modified in said version of said memory state committed in step (c), wherein one of said first and second uncommitted versions is aborted when the other of said first and second uncommitted versions is committed in step (c).

33. The method as claimed in claim 32, wherein said beginning preparation of results of said read-write transactions includes following a predefined strategy for selecting either said last committed version or one of said uncommitted versions for access by a third transaction for which preparation has begun.

34. The method as claimed in claim in claim 33, wherein said predefined strategy includes initially selecting said last committed version, and from said last committed version preparing a third uncommitted version associated with said third transaction, and after a conflict is detected between said third uncommitted version and said first uncommitted version wherein said third uncommitted version is based upon a resource of said last committed version having been modified in said first uncommitted version, preparing from said first uncommitted version a fourth uncommitted version associated with said third transaction.