Adaptive reader-writer lock

• US 7,430,627 B2
• Filed: 02/17/2005
• Issued: 09/30/2008
• Est. Priority Date: 12/19/2000
• Status: Expired due to Fees

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: McKenney, Paul E.
• Primary Examiner(s): Auve; Glenn A.
• Assistant Examiner(s): Vu; Trisha
• Application Number:
  US11/059,921
• Publication Number:
  US 20050149634A1
• Time in Patent Office: 1,321 Days
• Field of Search: 707/8, 709/229, 718/104, 718/107, 710/200
• US Class Current: 710/200
• CPC Class Codes: G06F 9/52 Program synchronisation; Mutual exclusion, e.g. by means of semaphores

Abstract

A method and computer system for dynamically selecting an optimal synchronization mechanism for a data structure in a multiprocessor environment. The method determines a quantity of read-side and write-side acquisitions, and evaluates the data to determine an optimal mode for efficiently operating the computer system while maintaining reduced overhead. The method incorporates data received from the individual units within a central processing system, the quantity of write-side acquisitions in the system, and data which has been subject to secondary measures, such as formatives of digital filters. The data subject to secondary measures includes, but is not limited to, a quantity of read-side acquisitions, a quantity of write-side acquisitions, and a quantity of read-hold durations. Based upon the individual unit data and the system-wide data, including the secondary measures, the operating system may select the most efficient synchronization mechanism from among the mechanisms available. Accordingly, efficiency of a computer system may be enhanced with the ability to selectively choose an optimal synchronization mechanism based upon selected and calculated parameters.

Claims

1. A method for dynamically determining a synchronization mechanism of a data structure in a multiprocessor, comprising:

maintaining first and second system-wide measures of read and write acquisitions;

determining said synchronization mechanism based upon at least some of said measures;

acquiring said synchronization mechanism for a first processing unit responsive to said determination; and

releasing said synchronization mechanism responsive to a change in at least some of said system-wide measures of read and write acquisitions, wherein the step of releasing said synchronization mechanism includes switching said first processing unit to a different synchronization mechanism.

2. The method of claim 1, wherein said synchronization mechanism is selected from the group consisting of:

a distributed reader-writer lock-, a centralized reader-writer lock, an exclusive lock, and RCU.

3. The method of claim 2, wherein said exclusive lock mode is selected from the group consisting of:

a test and set lock mode, a test and test and set lock mode, a queued lock mode, a ticket lock mode, and a quad-aware lock mode.

4. The method of claim 1, further comprising switching to a synchronization mechanism from another synchronization mechanism.

5. The method of claim 1, wherein the synchronization mechanism is a distributed reader-writer lock mode, and wherein said determining step is responsive to the system-wide measure of write acquisitions and the system-wide measure of read acquisitions.

6. The method of claim 5, wherein said determining step is further responsive to a quantity of units in the system.

7. The method of claim 6, wherein said unit is selected from a group consisting of:

a CPU, a thread, a processor, a transaction, a co-routine, a thread in a multi-threaded architecture, a NUMA module, and a task.

8. The method of claim 1, wherein the synchronization mechanism is a centralized lock mode, and wherein said determining step is responsive to the system-wide measure of write acquisitions and the system-wide measure of read acquisitions.

9. The method of claim 1, further comprising maintaining a system-wide measure of read-hold duration.

10. The method of claim 9, wherein the step of maintaining a system-wide measure of read-hold duration includes maintaining a measure of read-hold duration by a unit.

11. The method of claim 10, wherein said unit is selected from a group consisting of:

a CPU, a thread, a processor, a transaction, a co-routine, a thread in a multi-threaded architecture, a NUMA module, and a task.

12. The method of claim 9, wherein the synchronization mechanism is a centralized lock mode, and wherein said determining step is responsive to the system-wide measures of read acquisitions and read-hold duration.

13. The method of claim 9, wherein the synchronization mechanism is an exclusive lock mode and wherein said determining step is responsive to the system-wide measure of read-hold duration.

14. The method of claim 13, wherein said determining step is further responsive to the system-wide measure of read acquisitions.

15. The method of claim 1, further comprising periodically updating at least some of said system-wide measures.

16. The method of claim 1, wherein at least some of said second system-wide measures are selected from a group consisting of:

a digital filter, a weighted average, a sliding window average, a finite impulse response, and a central data structure.

17. A method for dynamically determining a synchronization mechanism of a data structure in a multiprocessor, comprising:

maintaining first and second system-wide measures of read and write acquisitions; and

determining said synchronization mechanism based upon at least some of said measures, wherein said synchronization mechanism is selected from the group consisting of;

a distributed reader-writer lock, a centralized reader-writer lock, an exclusive lock, and RCU.