Flexible acceleration of java thread synchronization on multiprocessor computers

• US 20030065704A1
• Filed: 09/28/2001
• Published: 04/03/2003
• Est. Priority Date: 09/28/2001
• Status: Active Grant

• Current Assignee: Intel Corporation
• Original Assignee: Intel Corporation
• Inventors: Buch, Deep K.
• Granted Patent: US 7,159,220 B2
• Time in Patent Office: Days
• US Class Current: 718/107
• CPC Class Codes: G06F 9/52 Program synchronisation; Mutual exclusion, e.g. by means of semaphores

Abstract

A method and machine-readable medium measure requests by threads requesting a lock to differentiate “hot” and “cold” locks in accordance with the level of contention for the locks. A hardware accelerator manages access to hot locks to improve performance.

Claims

1. A method comprising:

utilizing a monitor to control access by competing threads to a shared resource,

said monitor having a data structure to process lock requests;

providing an extension in said monitor data structure to keep a count of unique thread accesses in a given time period; and

registering a number of unique requests for a given lock by contending threads and registering a count indicative of the number of contending threads in the given time period.

2. The method of claim 1 wherein producing a count of contended threads comprises sensing a thread request for a monitor, determining if the requesting thread is different from a previous requesting thread and incrementing a monitor access count indicative of contentions in response to a comparison indicating that a requesting thread is different from a previously requesting thread.

3. The method according to claim 2 comprising comparing the monitor access count with a threshold value, determining when the count exceeds a threshold, and identifying a lock as “

being highly contended”

in response to the count exceeding the threshold.

4. The method according to claim 3 further comprising responding to the identification of a lock as being highly contended for by selecting a lock request in accordance with a method selected for highly contended for locks.

5. The method according to claim 4 wherein utilization of hardware acceleration to process lock requests comprises a pre-selected number of processors each issuing requests for a given lock to a shared resource, arbitrating among requests, notifying a current winner processor providing access;

issuing to a processor not gaining access a special flag variable in cacheable memory, modifying the flag variable in memory to grant access to the processor that did not gain access when a prior processor releases its lock.

6. A machine-readable medium that provides instructions, which when executed by a processor, causes said processor to perform operations comprising:

utilizing a monitor to control access by competing threads to a shared resource, said monitor having a data structure to process lock requests;

providing an extension in said monitor data structure to keep a count of unique thread accesses in a given time period; and

registering a number of unique requests for a given lock by contending threads and registering a count indicative of the number of contending threads in the given time period.

7. The machine-readable medium in accordance with claim 6 wherein the instruction to register a count of contending threads comprises sensing a thread request for a monitor, determining if the requesting thread is different from a previous requesting thread and incrementing a monitor access count indicative of contentions in response to a comparison indicating that a requesting thread is different from a previously requesting thread.

8. The machine-readable medium according to claim 7 including an instruction of comparing the monitor access count with a threshold value, determining when the count exceeds a threshold, and identifying a lock as “

being highly contended”

in response to the count exceeding the threshold.

9. The machine-readable medium according to claim 8 further comprising an instruction of responding to the identification of a lock as being highly contended by selecting a lock request in accordance with a method selected for highly contended for locks.

10. The machine-readable medium according to claim 6 wherein the instruction to utilize hardware acceleration to process lock requests comprises a pre-selected number of processors each issuing requests for a given lock to a shared resource, arbitrating among requests, notifying a current winner processor and providing access;

and

as to a processor not gaining access, adding a special flag variable in cacheable memory, modifying the flag variable in memory to grant access to the processor that did not gain access when a prior processor releases its lock.

11. A machine-readable medium according to claim 10 wherein said machine-readable medium is a Java virtual machine.

12. A lock register comprising a processor lock register to register access requests by each of the plurality of processors to a shared resource and a lock current ownership data structure corresponding to each of said locks, said lock data structure registering identity of said lock, identity of contending threads and identity of a processor on which said thread is executing.

13. The lock register according to claim 12 further comprising a spin flag field, said lock register further comprising an arbiter to select a processor for access to a lock and to issue a spin flag to a contending processor not awarded connection to the lock.

14. The lock register according to claim 13 wherein issuing a spin flag to said processor not gaining access further comprises assigning access priority of the contending processor for connection when a previous processor gives up control of a lock.

15. The lock register according to claim 14 comprising a register field for sensing when no registers are available.