Counter and flux bit locking for very fast shared serialization of shared data objects

• US 5,293,600 A
• Filed: 04/06/1990
• Issued: 03/08/1994
• Est. Priority Date: 04/06/1990
• Status: Expired due to Fees

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Vriezen, John J.
• Primary Examiner(s): Kulik, Paul V.
• Assistant Examiner(s): LINTZ, PAUL R
• Application Number:
  US07/505,814
• Time in Patent Office: 1,432 Days
• Field of Search: 364/DIG. 1, 364/DIG. 2, 395/425, 395/325, 395/775, 395/600
• US Class Current: 711/152
• CPC Class Codes: G06F 9/52 Program synchronisation; Mutual exclusion, e.g. by means of semaphores
• CPC Class Codes: Y10S 707/99938 Concurrency, e.g. lock management in shared database

Abstract

Methods and apparatus are set forth for improving the performance of general purpose computing systems that utilize exclusive and shared seize mechanisms to assure the integrity of shared data objects, records, or other unit of synchronization. More particularly, methods and apparatus are set forth which allow concurrent reads of shared data objects in a coherent fashion, without always having to perform a seize. According to a preferred embodiment of the invention, read (shared) accesses can be performed in such a manner as to improve system performance by utilizing a method comprising the steps of: (a) determining, for a particular shared data object to be read, if the object is subject to an exclusive seize prior to any attempt, to read the object; (b) reading the object without causing a shared seize to be initiated if it was not subject to an exclusive seize as determined in step (a); (c) determining if the object is subject to an exclusive seize after reading it in accordance with step (b); and (d) determining if the object was subject to an exclusive seize while it was being read. If, as a result of performing the aforestated method, it can be concluded that the read took place without an exclusive seize immediately before, during and just after the read, then the data can be presumed valid. System performance is improved in such cases since the overhead associated with initiating a seize for the read is eliminated.

Claims

1. A method for managing access to shared data in a data processing system, comprising the steps of:

(a) initializing an n bit check count to keep track of each instance where an exclusive write access modifies a given shared data object;

(b) storing said check count in a has table; and

(c) setting a flux bit whenever said data object is subject to an exclusive seize, so that any shared read access initiated while said flux bit is set will initiate a shared seize prior to reading the data object.

2. A method as set forth in claim 1 further comprising the steps of:

(a) incrementing said check count each time an exclusive write access causes said data object to be modified;

(b) accessing said check count both before and after reading said given data object when performing a shared read access with respect to said data object, to determine if an exclusive access modified said data object during the performance of said shared read access;

(c) disregarding the shared read if the accessed check count values accesses in step (b) are not equal; and

(d) completing said shared read access without having performed a shared seize whenever said check count values are equal.

3. A method for improving the performance of a computing system that utilizes exclusive and shared seize mechanisms to assure the integrity of shared data objects, comprising the steps of:

(a) initiating a shared read for a particular shared data object;

(b) determining if said object is subject to an exclusive seize prior to any attempt to read the object;

(c) reading said object without initiating a shared seize whenever the object is not subject to an exclusive seize as determined in step (b);

(d) determining if the object is subject to an exclusive seize after performing step (c); and

(e) determining, whenever said object is not subject to an exclusive seize as determined in step (d), if the object was subject to an exclusive seize while being read; and

(f) completing said shared read without having initiated a shared seize, thereby improving system performance, whenever it is determined by the performance of steps (b), (d) and (e) that no exclusive seize affected the validity of said shared data object since the initiation of said read.

4. A method as set forth in claim 3 wherein said step (b) is performed by examining the condition of a flux bit associated with said object, wherein said flux bit is set whenever said object is subject to an exclusive seize.

5. A method as set forth in claim 3 wherein said step (d) is performed by examining the condition of a flux bit associated with said object, wherein said flux bit is reset whenever said object is not subject to an exclusive seize.

6. A method as set forth in claim 3 wherein said step (e) is performed by maintaining a check count that is incremented whenever said shared data object is subject to an exclusive seize and modified by an exclusive write access.

7. A method as set forth in claim 4 wherein said check count is stored in a hash table.

8. A method as set forth in claim 6 wherein steps (b)-(e) are performed iteratively whenever it is determined by the performance of step (e) that said object was modified by an exclusive write access while being read.

9. A method as set forth in claim 8 wherein the number of iterative performances 94 steps (b)-(e) is limited by a predefined retry threshold value.

10. A method as set forth in claim 9 further comprising the step of performing a shared seize whenever said predefined threshold value is met.

11. A method for managing access to shared data in a data processing system, wherein said data processing system utilizes exclusive and shared seize mechanisms to assure the integrity of shared data objects, comprising the steps of:

(a) initiating an exclusive write for a particular shared data object, including initiating an exclusive seize of said object;

(b) setting a flux bit associated with said object whenever an exclusive seize is acquired;

(c) initializing and maintaining an n bit check count for keeping track of each instance where said data object is modified by an exclusive write access;

(d) modifying said given data object via said exclusive write;

(e) incrementing said check count after said data is modified via said exclusive write;

(f) resetting said flux bit when said object is no longer subject to said exclusive seize; and

(g) terminating said exclusive write.

12. A method as set forth in claim 11 further comprising the steps of:

(a) determining and storing the value of said check count after a shared read operation is initiated but prior to reading said data object, said shared read operation being initiated to obtain data;

(b) determining and storing the value of said check count after reading said data object but prior to completing said shared read operation;

(c) comparing the check count value determined and stored in accordance with steps (a)-(b); and

(d) invalidating said data whenever said check count values are different.

13. A method as set forth in claim 12 further comprising the step of retrying said shared read whenever said check count values are different.

14. A method as set forth in claim 13 wherein said step of retrying is performed only so long as said check counts differ and a predetermined retry threshold value is not met.

15. A method as set forth in claim 14 further comprising the step of initiating a shared seize whenever said predetermined retry threshold is met.

16. A method as set forth in claim 14 further comprising the step of completing said shared read, without initiating a shared seize, so long as said flux bit remains reset during the performance of said shared read and said check count values are the same.

17. A method as set forth in claim 12 further comprising the step of initiating a shared seize whenever it is determined during the performance of a shared read that said flux bit is set.

18. A method as set forth in claim 12 further comprising the step of retrying said shared read if it is determined during the performance of said shared read that said flux bit is set.

19. A method as set forth in claim 11 wherein the value of n is such that there does not exist a one to one correspondence between the number of shared data objects in said system and the maximum possible number of check count values, and further wherein said check count values are stored in a table accessible via a hash function mapping at least one shared data object into a given table entry.

20. A method for managing access to shared data in a computing system in which a plurality of tasks may be executed concurrently utilizing a shared data base, wherein said computing system includes a lock manager for performing read (shared) accesses and write (exclusive) accesses with respect to shared data objects in said data base through the use of exclusive and shared seize mechanisms, comprising the steps of:

(a) initializing a counter to keep track of shared read attempts with respect to a particular shared data object whenever a shared read access in initiated for that object;

(b) incrementing said counter;

(c) initializing a first check count variable by setting it equal to a check count associated with said particular data object, wherein said check count is a variable that is incremented whenever a modification is made to said particular data object under an exclusive seize;

(d) testing a flux bit which is set whenever said particular data object is subject to an exclusive seize;

(e) reading said particular shared data object so long as a flux bit is reset and the value in said counter after performing step (b) is less than a predetermined threshold value;

(f) determining if said particular data object is subject to an exclusive seize after performing step (e) by again testing said flux bit;

(g) determining, whenever said particular data object is not subject to an exclusive seize as determined in step (f), if said particular data object was subject to an exclusive seize while it was being read; and

(h) terminating said shared read access successfully without having initiated a shared seize whenever it is determined in step (g) that said particular data object was not subject to an exclusive seize while it was being read.

21. A method for improving the performance of a computer system utilizing exclusive and shared seize mechanisms to assure the integrity of shared data objects, comprising the steps of:

(a) setting a flux bit whenever an exclusive write access is in progress and a shared data object is subject to an exclusive seize;

(b) incrementing a check count whenever said shared data object is modified during said exclusive write access;

(c) resetting said flux bit whenever said shared data object is no longer subject to said exclusive seize;

(d) testing said flux bit whenever a shared read access for said shared data object is initiated; and

(e) performing a shared seize with respect to said shared data object whenever a shared read access is initiated for said shared data object and said flux bit is set.

22. A method as set forth in claim 21 further comprising the steps of:

(a) reading said shared data object without performing a shared seize, whenever said flux bit is not set;

(b) testing said flux bit after performing said read; and

(c) performing a shared seize with respect to said shared data object whenever said flux bit is set after performing said read.

23. A method as set forth in claim 22 further comprising the steps of:

(a) storing the value of said check count in a first storage means prior to performing said read;

(b) storing the value of said check count in a second storage means after performing said read so long as said flux bit is not set after said read;

(c) comparing the value of said check count as stored in said first and second storage means;

(d) disregarding said read if said check count values as stored in said first and second storage means are not equal; and

(e) completing said shared read access without having performed a share seize whenever said check count values as stored in said first and second storage means are equal.

24. A method as set forth in claim 23 further comprising the steps of:

(a) retrying said shared read access whenever said read is disregarded so long as a predetermined retry threshold is not exceeded and said flux bit remains reset; and

(b) performing a shared seize for said data object when performing said shared read access whenever said retry threshold is exceeded or said flux bit is set.

25. A method for improving the performance of a computer system utilizing exclusive and shared seize mechanism to assure the integrity of shared data objects, comprising the steps of:

(a) maintaining a check count in a hash table for each of a plurality of shared data objects, wherein the address of the check count in said table for a given shared data object is obtained via performing a hash function on the address in said system of said shared data object;

(b) utilizing said check counts to determine if a shared read access with respect to said given object can be performed without having to seize said object;

(c) incrementing said check count whenever said shared data object is modified by an exclusive write access while subject to an exclusive seize;

(d) setting a flux bit whenever said shared data object is subject to an exclusive seize; and

(e) resetting said flux bit whenever said shared data object is no longer subject to said exclusive seize.

26. A method as set forth in claim 25 further comprising the step of forcing a shared seize of said data object whenever a shared read is initiated and said flux bit for said data object is set.

27. A method as set forth in claim 25 further comprising the step of forcing a shared seize of said data object after a predetermined number of attempts to perform said shared read without having to perform said shared seize.

28. A method as set forth in claim 27 further comprising the step of completing said shared read without having initiated a shared seize so long as the flux bit for said object is not set after said shared read is initiated and said check count indicates that said shared data object was not modified during said read.

29. An apparatus for improving the performance of a computing system that utilizes exclusive and shared seize mechanisms to assure the integrity of shared data objects, said apparatus comprising:

(a) means for initiating a shared read for a particular shared data object;

(b) means for determining if said object is subject to an exclusive seize prior to any attempt to read the object;

(c) means for reading said object without initiating a shared seize whenever the object is not subject to an exclusive seize;

(d) means for determining if the object is subject to an exclusive seize after said object has been read;

(e) means for determining if the object is subject to an exclusive seize while being read; and

(f) means for completing said shared read without initiating a shared seize to improve system performance if no exclusive seize has been determined.

30. The apparatus as claimed in claim 29 wherein said means for determining if said object is subject to an exclusive seize prior to any attempt to read the object includes a flux bit associated with said object and means for examining the condition of said flux bit to determine if said object is subject to an exclusive seize, said flux bit being set whenever said object is subject to an exclusive seize.

31. The apparatus as claimed in claim 29 wherein said means for determining if the object is subject to an exclusive seize after said object has been read includes a flux bit associated with said object and means for examining the condition of said flux bit, wherein said flux bit is reset whenever said object is not subject to an exclusive seize.

32. The apparatus as claimed in claim 29 wherein said means for completing includes means for maintaining a check count and incrementing said check count whenever said shared data object is subject to an exclusive seize.

33. An apparatus for managing access to shared data in a data processing system including a computing system utilizing exclusive and shared seize mechanisms to assure the integrity of shared data objects, said apparatus comprising:

(a) means for initiating an exclusive write for a particular shared object, said means for initiating including means for initiating an exclusive seize of said objects;

(b) a flux bit associated with said object and means for setting said flux bit whenever an exclusive seize is acquired;

(c) means for initializing and maintaining an n bit check count for keeping track of each instance where said data object is modified by an exclusive write access;

(d) means for modifying said data object via said exclusive write;

(e) means for incrementing said check count after said data is modified via said exclusive write;

(f) means for resetting said flux bit when said object is no longer subject to said exclusive seize; and

(g) means for terminating said exclusive write.

34. The apparatus as set forth in claim 33 further comprising:

(a) means for determining and storing the value of said check count after a shared read is initiated but prior to reading said data objects;

(b) means for determining and storing the value of said check count after reading said data object but prior to completing said shared read;

(c) means for comparing said determined check count values; and

(d) means for invalidating the read of said object whenever said determined check count values are different from each other.

35. An apparatus for managing access to shared data objects in a computing system in which a plurality of tasks may be executed concurrently utilizing a shared data base, wherein said computing system includes a lock manager for performing read (shared) accesses and write (exclusive) accesses with respect to said shared data objects in said base through the use of exclusive and shared seize mechanisms, said apparatus comprising;

(a) means for initiating and acquiring an exclusive seize via said lock manager with respect to a given data object whenever an exclusive access with respect to said given data object is initiated by one of said plurality of tasks;

(b) a flux bit and means for setting said flux bit after said seize is acquired;

(c) means for modifying said given data object after said flux bit has been set;

(d) means for incrementing a check count to signify that a modification of said given data object has occurred;

(e) means for resetting said flux bit; and

(f) means for releasing said exclusive seize after said flux bit has been reset.

36. An apparatus for managing access to shared data in a computing system in which a plurality of tasks may be executed concurrently utilizing a shared data base, wherein said computing system includes a lock manager for performing read (shared) accesses and write (exclusive) accesses with respect to shared data objects in said data base through the use of exclusive and shared seize mechanisms, said apparatus comprising:

(a) a counter and means for initializing said counter to keep track of shared read attempts with respect to a particular shared data object whenever a shared read access is initiated for that object;

(b) means for incrementing said counter;

(c) means for initializing a first check count variable by setting it equal to check count associated with said particular data object, wherein said means for initializing includes means for incrementing said check count variable whenever a modification is made to said particular data object under an exclusive seize;

(d) a flux bit which is set whenever said particular data object is subject to an exclusive seize, and means for testing said flux bit;

(e) means for reading said particular shared data object when said flux bit is reset and the value in said counter after said counter is incremented is less than a predetermined threshold value;

(f) first means for determining if said particular data object is subject to an exclusive seize after said particular data object is read, said first means for determining including means for again testing said flux bit;

(g) second means for determining whenever said particular data object is not subject to an exclusive seize after again testing said flux bit, if said particular data object was subject to an exclusive seize while it was being read; and

(h) means for completing said shared read access without initiating a shared seize whenever it is determined that said particular data object was not subject to an exclusive seize while it was being read.

37. An apparatus for improving the performance of a computer system utilizing exclusive and shared seize mechanisms to assure the integrity of shared data objects, said apparatus comprising;

(a) a flux bit and means for setting said flux bit whenever an exclusive write access is in progress and a shared data object is subject to an exclusive seize;

(b) means for incrementing a check count whenever said shared data object is modified during said exclusive write access;

(c) means for resetting said flux bit whenever said shared data object is no longer subject to said exclusive seize;

(d) means for testing said flux bit whenever a shared read access for said shared data object is initiated; and

(e) means for performing a shared seize with respect to said shared data object whenever a shared read access is initiated for said shared data object and said flux bit is set.