Transactional Memory Computing System with Support for Chained Transactions

• US 20090113443A1
• Filed: 10/30/2007
• Published: 04/30/2009
• Est. Priority Date: 05/14/2007
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Baum, Richard J., Heller Jr., Thomas J.
• Granted Patent: US 8,095,741 B2
• Time in Patent Office: Days
• US Class Current: 718/106
• CPC Class Codes: G06F 11/141 for bus or memory accesses
• CPC Class Codes: G06F 9/3004 to perform operations on memory
• CPC Class Codes: G06F 9/30087 Synchronisation or serialisation instructions

Abstract

A computing system processes memory transactions for parallel processing of multiple threads of execution provides execution of multiple atomic instruction groups (AIGs) on multiple systems to support a single large transaction that requires operations on multiple threads of execution and/or on multiple systems connected by a network. The support provides a Transaction Table in memory and fast detection of potential conflicts between multiple transactions. Special instructions may mark the boundaries of a transaction and identify memory locations applicable to a transaction. A ‘private to transaction’ (PTRAN) tag, directly addressable as part of the main data storage memory location, enables a quick detection of potential conflicts with other transactions that are concurrently executing on another thread. The tag indicates whether (or not) a data entry in memory is part of a speculative memory state of an uncommitted transaction that is currently active in the system.

Claims

1. A method for use by a transaction program on a computing system for managing memory access to a shared memory location for transaction data of a first thread, the shared memory location being accessible by the first thread and a second thread, the method comprising:

using multiple atomic instruction groups (AIGs) of instruction to support on multiple systems excution of a single large transaction that requires operations on multiple threads, and tracking read and write sets of chained AIGs individually.

2. The method according to claim 1 wherein said multiple threads are executed on multiple systems connected by a network.

3. The method according to claim 1 wherein for excution of chained AIGs, a determination is made whether an instruction which is part of an active atomic instruction group (AIG) of instructions associated with a transaction may require the services of another thread of execution on said computing system or in another system, and when an originating AIG includes a message to be sent to another system, then generating a messaging request to a specialized service API and to a service provider that has been changed to make use of an AIG for speculative execution, said message request indicating that the request is part of a speculative execution which has not been committed to memory.

4. The method according to claim 3 wherein when said service provider receives the message request it begins a chained AIG on its own thread and establishes a chain of AIGs which may be rolled back in case of a transaction failure.

5. The method according to claim 4 wherein said service provider executes said chained AIG it begins on its own thread which includes code to execute a service for speculative execution, and response to the specialized service API is made when the service is complete and ready to commit.

6. The method according to claim 4 wherein the chained AIG is not committed until the originating AIG is also ready to commit.

7. The method according to claim 4 wherein the commit code for the originating AIG is modified to support a chain of AIGs that has been established by the specialized service API.

8. The method according to claim 7 wherein when a chain is known to span multiple threads of executions on multiple systems all participants in the chanin are notified when an originating AIG should be committed.

9. The method according to claim 7 wherein a call to a specialized service API is included in the original AIG.

10. The method according to claim 3 wherein the computing system provides multiple transaction tables enabling the individual tracking and read and write sets of chained AIGs and the rolling of them back individually when needed.

11. A computer-readable computer program medium tangibly embodying computer-executable instructions for performing a computer process implementing a string of instructions to complete a transaction of a first thread among multiple threads, beginning with one instruction of said string of instructions, and execution said string of instructions, and having one instruction set which is part of an active atomic instruction group (AIG) of instructions having a call to a specialized service enabling using multiple atomic instruction groups (AIGs) of instruction to support on multiple systems excution of a single large transaction that requires operations on multiple threads, and tracking read and write sets of chained AIGs individually.

12. A computing system, comprising:

a plurality of processors and a shared memory accessible by one or more of said processors while said computing system manages memory access to a shared memory location for a transaction being processed, said shared memory location being accessible by a first thread and a second thread,

said computing system shared memory having an identified ‘

private to transaction’

tag for identifying each increment of system memory and a transaction table of loads and stores of a first thread for said transaction being processed, and having support for execution on multiple processors of a single large transaction that requires operations on multiple threads using chained AIGs, and tracking read and write sets of chained AIGs individually.

13. The computing system according to claim 12 wherein a call to a specialized service API is included an original AIG to initiate said support for execution on multiple processors of said single large transaction using chained AIGs.

14. The computing system according to claim 12 wherein said multiple threads are executed on multiple processors connected by a network.

15. The method according to claim 12 wherein for excution of chained AIGs, a determination is made whether an instruction which is part of an active atomic instruction group (AIG) of instructions associated with a transaction may require the services of another thread of execution on a processor of said computing system or in another system, and when an originating AIG includes a message to be sent to another system, then generating a messaging request to a specialized service API and to a service provider that has been changed to make use of an AIG for speculative execution, said message request indicating that the request is part of a speculative execution which has not been committed to memory.

16. The method according to claim 14 wherein when said service provider receives the message request it begins a chained AIG on its own thread and establishes a chain of AIGs which may be rolled back in case of a transaction failure, and wherein multiple tables enable the individual tracking and read and write sets of chained AIGs and the rolling of them back individually when needed.

17. The method according to claim 15 wherein said service provider executes said chained AIG it begins on its own thread which includes code to execute a service for speculative execution, and response to the specialized service API is made when the service is complete and ready to commit.

18. The method according to claim 15 wherein the chained AIG is not committed until the originating AIG is also ready to commit.

19. The method according to claim 15 wherein the commit code for the originating AIG is modified to support a chain of AIGs that has been established by the specialized service API.

20. The method according to claim 19 wherein when a chain is known to span multiple threads of executions on multiple systems all participants in the chanin are notified when an originating AIG should be committed.