Method, apparatus, and computer program product for dynamically tuning a data processing system by identifying and boosting holders of contentious locks

• US 7,458,076 B2
• Filed: 08/05/2004
• Issued: 11/25/2008
• Est. Priority Date: 08/05/2004
• Status: Expired due to Fees

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Swanberg, Randal Craig, Warrier, Suresh Eswara
• Primary Examiner(s): An; Meng-Ai
• Assistant Examiner(s): ZHE, MENG YAO
• Application Number:
  US10/912,500
• Publication Number:
  US 20060031658A1
• Time in Patent Office: 1,573 Days
• Field of Search: 718/100
• US Class Current: 718/100
• CPC Class Codes: G06F 9/3851 from multiple instruction streams, e.g. multistreaming
• CPC Class Codes: G06F 9/3885 using a plurality of independent parallel functional units
• CPC Class Codes: G06F 9/485 Task life-cycle, e.g. stopping, restarting, resuming execution G06F9/4881 takes precedence
• CPC Class Codes: G06F 9/4881 Scheduling strategies for dispatcher, e.g. round robin, multi-level priority queues
• CPC Class Codes: G06F 9/526 Mutual exclusion algorithms

Abstract

A method, apparatus, and computer program product are disclosed for a simultaneous multithreading (SMT) data processing system for modifying the processing of software threads that acquire a contentious software lock. The system includes a processor that is capable of concurrently executing multiple different threads on the processor. The processor is also capable of utilizing hardware thread priorities assigned to each thread the processor is processing by granting a greater, disparate amount of resources to the highest priority thread. A hardware priority is assigned to each one of the SMT threads. A contentious lock is identified. Ones of the multiple threads are identified that attempt to acquire the contentious lock. These threads are dynamically redirected to special code for handling contentious locks. The hardware priority of a thread acquiring a contentious lock is then boosted. According to the preferred embodiment, the present invention redirects callers of a locking function to a “hot” locking function when the lock they are acquiring has been classified as “hot”, and vice versa.

Claims

1. A method in a simultaneous multithreading data processing system for modifying the processing of multiple software threads that acquire a software lock, said method comprising:

concurrently executing multiple software threads on a processor included in said data processing system;

assigning a hardware priority to each one of said multiple software threads;

identifying the software lock as a contentious lock by a first locking function;

boosting, by the first locking function, a hardware priority of each of said multiple software threads that successfully acquires said contentious lock;

identifying a command that was executed by one of the multiple software threads to branch to a second locking function;

tracking, by said second locking function, a number of times said one of the multiple software threads successfully acquires a lock on said first attempt to acquire said lock;

comparing, by said second locking function, said number of times to a non-contention threshold;

in response to said number of times exceeding said non-contention threshold, identifying said lock as a non-contentious lock;

in response to identifying said lock as the non-contentious lock, storing an address of said first locking function in said command; and

in response to a next execution of said command, branching to said first locking function instead of branching to said second locking function and continue executing the multiple software threads.

2. The method according to claim 1, further comprising:

tracking, by said first locking function, a number of times one of said threads attempts unsuccessfully to acquire a lock;

comparing, by said first locking function, said number of times to a contention threshold; and

in response to said number of times exceeding said contention threshold, identifying said lock as a contentious lock.

3. The method according to claim 2, further comprising:

identifying a command that was executed by said thread to branch to said first locking function;

in response to identifying said lock as a contentious lock, storing an address of a second locking function in said command; and

in response to a next execution of said command, branching to said second locking function instead of branching to said first locking function.

4. The method according to claim 1, further comprising:

dynamically adjusting said contention threshold during runtime of said system; and

dynamically adjusting said non-contention threshold during runtime of said system.

10. The product according to claim 1, further comprising:

instructions for tracking, by said first locking function, a number of times one of said threads attempts unsuccessfully to acquire a lock;

instructions for comparing, by said first locking function, said number of times to a contention threshold; and

in response to said number of times exceeding said contention threshold, instructions for identifying said lock as a contentious lock.

11. The product according to claim 10, further comprising:

instructions for identifying a command that was executed by said thread to branch to said first locking function;

in response to identifying said lock as a contentious lock, instructions for storing an address of a second locking function in said command; and

in response to a next execution of said command, instructions for branching to said second locking function instead of branching to said first locking function.

5. An apparatus in a simultaneous multithreading data processing system for modifying the processing of multiple software threads that acquire a software lock, said apparatus comprising:

a processor concurrently executing multiple software threads;

a hardware priority assigned to each one of said multiple software threads;

a first locking function for identifying the software lock as a contentious lock;

said first locking function boosting a hardware priority of each of said multiple software threads that successfully acquires said contentious lock;

a second locking function for identifying a command that was executed by one of the multiple software threads to branch to a second locking function;

said second locking function tracking a number of times said one of the multiple software threads successfully acquires a lock on said first attempt to acquire said lock;

said second locking function comparing said number of times to a non-contention threshold;

in response to said number of times exceeding said non-contention threshold, said second locking function identifying said lock as a non-contentious lock;

in response to identifying said lock as a non-contentious lock, said second locking function storing an address of said first locking function in said command; and

in response to a next execution of said command, said processor branching to said first locking function instead of branching to said second locking function.

6. The apparatus according to claim 5, further comprising:

said first locking function tracking a number of times one of said threads attempts unsuccessfully to acquire a lock;

said first locking function comparing said number of times to a contention threshold; and

in response to said number of times exceeding said contention threshold, said first locking function identifying said lock as a contentious lock.

7. The apparatus according to claim 6, further comprising:

said first locking function identifying a command that was executed by said thread to branch to said first locking function;

in response to identifying said lock as a contentious lock, said first locking function storing an address of a second locking function in said command; and

in response to a next execution of said command, said processor branching to said second locking function instead of branching to said first locking function.

8. The apparatus according to claim 5, further comprising:

said contention threshold being dynamically adjusted during runtime of said system; and

said non-contention threshold being dynamically adjusted during runtime of said system.

9. A computer program product stored in a recordable-type media of simultaneous multithreading data processing system for modifying the processing of multiple software threads that acquire a software lock, said computer program product comprising:

instructions for concurrently executing multiple software threads on a processor included in said data processing system;

instructions for assigning a hardware priority to each one of said multiple software threads;

instructions for identifying the software lock as a contentious lock by a first locking function;

instructions for boosting, by the first locking function, a hardware priority of each of said multiple software threads that successfully acquires said contentious lock;

instructions for identifying a command that was executed by one of the multiple software threads to branch to a second locking function;

instructions for tracking, by said second locking function, a number of times said one of the multiple software threads successfully acquires a lock on said first attempt to acquire said lock;

instructions for comparing, by said second locking function, said number of times to a non-contention threshold;

in response to said number of times exceeding said non-contention threshold, instructions for identifying said lock as a non-contentious lock;

in response to identifying said lock as a non-contentious lock, instructions for storing an address of said first locking function in said command; and

in response to a next execution of said command, instructions for branching to said first locking function instead of branching to said second locking function.

12. The product according to claim 9, further comprising:

instructions for dynamically adjusting said contention threshold during runtime of said system; and

instructions for dynamically adjusting said non-contention threshold during runtime of said system.