Mutex profiling based on waiting analytics

• US 10,417,057 B2
• Filed: 01/30/2017
• Issued: 09/17/2019
• Est. Priority Date: 01/30/2017
• Status: Active Grant

• Current Assignee: Oracle International Corporation (Oracle Corporation)
• Original Assignee: Oracle International Corporation (Oracle Corporation)
• Inventors: Loh, Eugene Ying
• Primary Examiner(s): An, Meng Ai T
• Assistant Examiner(s): Chu Joy-Davila, Jorge A
• Application Number:
  US15/419,699
• Publication Number:
  US 20180217878A1
• Time in Patent Office: 960 Days
• CPC Class Codes: G06F 9/526 Mutual exclusion algorithms

Abstract

Techniques are described herein for profiling mutual exclusion locks. In one or more embodiments, a performance profiler identifies a first thread executing an activity within a parallel processing environment. The performance profiler identifies a set of one or more mutual-exclusion locks held by the first thread while the activity is being executed and determines the total number of threads waiting for at least one of the set of mutual-exclusion locks held by the first thread. The performance profiler then determines a total cost associated with execution of the activity based at least on the total number of threads waiting for at least one of the set of mutual-exclusion locks held by the first thread during execution of the activity. In one or more embodiments, the total cost is obtained by multiplying a raw cost with the number of threads waiting.

Claims

1. A non-transitory computer-readable medium comprising instructions which, when executed by one or more hardware processors, causes performance of operations comprising:

identifying a first thread executing an activity;

determining a set of one or more mutual-exclusion locks held by the first thread while the activity is being executed;

determining a total number of threads waiting for at least one of the set of one or more mutual-exclusion locks held by the first thread during execution of the activity;

stochastically interrupting the first thread executing the activity to determine a set of one or more values representing costs associated with executing the activity;

generating a profile for the activity based at least in part on the set of one or more values representing costs associated with executing the activity;

determining a total cost associated with execution of the activity as a function of at least the set of one or more values representing at least one cost associated with the first thread executing the activity and the total number of threads waiting for at least one of the set of one or more mutual-exclusion locks held by the first thread during execution of the activity; and

implementing, based at least in part on the total cost associated with execution of the activity, a performance optimization by modifying code for executing the activity to reduce mutex wait times caused by execution of the activity.

2. The non-transitory computer-readable medium of claim 1, wherein determining the total cost associated with execution of the activity comprises:

multiplying a value of the one or more values representing a raw cost associated with execution of the activity by the total number of threads waiting for at least one of the set of one or more mutual-exclusion locks.

3. The non-transitory computer-readable medium of claim 1, wherein determining the total cost associated with execution of the activity comprises:

multiplying a value of the one or more values representing a period of time for completing execution of the activity by the total number of threads waiting for at least one of the set of one or more mutual-exclusion locks.

4. The non-transitory computer-readable medium of claim 1, wherein determining the total cost associated with execution of the activity comprises:

multiplying a value of the one or more values representing a cost associated with a cache miss with a total number of threads waiting for at least one of the set of one or more mutual-exclusion locks held by the first thread during execution of the activity.

5. The non-transitory computer-readable medium of claim 1, wherein the operations further comprise:

incrementing a counter associated with a particular mutual-exclusion lock of the set of one or more mutual-exclusion locks in response to a particular thread waiting for the particular mutual-exclusion lock, wherein the counter identifies a number of threads currently waiting for the particular mutual-exclusion lock;

decrementing the counter in response to the particular thread obtaining the mutual-exclusion lock.

6. The non-transitory computer-readable medium of claim 1, wherein implementing the performance optimization comprises modifying the code to defer execution of a calculation until after release of at least one of the set of one or more mutual-exclusion locks or perform execution of the calculation before acquiring at least one of the set of one or more mutual-exclusion locks.

7. The non-transitory computer-readable medium of claim 1, wherein implementing the performance optimization comprises modifying a granularity corresponding to at least one of the set of one or more mutual-exclusion locks.

8. The non-transitory computer-readable medium of claim 1, wherein implementing the performance optimization comprises modifying a global variable in the code to a local variable.

9. The non-transitory computer-readable medium of claim 1, wherein determining the total cost associated with execution of the activity is further based on a cost associated with execution of the activity by a second thread.

10. A method comprising:

identifying a first thread executing an activity;

determining a set of one or more mutual-exclusion locks held by the first thread while the activity is being executed;

determining a total number of threads waiting for at least one of the set of one or more mutual-exclusion locks held by the first thread during execution of the activity;

stochastically interrupting the first thread executing the activity to determine a set of one or more values representing costs associated with executing the activity;

generating a profile for the activity based at least in part on the set of one or more values representing costs associated with executing the activity;

determining a total cost associated with execution of the activity as a function of at least the set of one or more values representing at least one cost associated with the first thread executing the activity and the total number of threads waiting for at least one of the set of one or more mutual-exclusion locks held by the first thread during execution of the activity; and

implementing, based at least in part on the total cost associated with execution of the activity, a performance optimization by modifying code for executing the activity to reduce mutex wait times caused by execution of the activity.

11. The method of claim 10, wherein determining the total cost associated with execution of the activity comprises:

multiplying a value of the one or more values representing a raw cost associated with execution of the activity by the total number of threads waiting for at least one of the set of one or more mutual-exclusion locks.

12. The method of claim 10, wherein determining the total cost associated with execution of the activity comprises:

multiplying a value of the one or more values representing a period of time for completing execution of the activity by the total number of threads waiting for at least one of the set of one or more mutual-exclusion locks.

13. The method of claim 10, wherein determining the total cost associated with execution of the activity comprises:

multiplying a value of the one or more values representing a cost associated with a cache miss with a total number of threads waiting for at least one of the set of one or more mutual-exclusion locks held by the first thread during execution of the activity.

14. The method of claim 10, further comprising:

incrementing a counter associated with a particular mutual-exclusion lock of the set of one or more mutual-exclusion locks in response to a particular thread waiting for the particular mutual-exclusion lock, wherein the counter identifies a number of threads currently waiting for the particular mutual-exclusion lock;

decrementing the counter in response to the particular thread obtaining the mutual-exclusion lock.

15. The method of claim 10, wherein implementing the performance optimization comprises modifying the code to defer execution of a calculation until after release of at least one of the set of one or more mutual-exclusion locks or perform execution of the calculation before acquiring at least one of the set of one or more mutual-exclusion locks.

16. The method of claim 10, wherein implementing the performance optimization comprises modifying a granularity corresponding to at least one of the set of one or more mutual-exclusion locks.

17. The method of claim 10, wherein implementing the performance optimization comprises modifying a global variable in the code to a local variable.

18. The method of claim 10, wherein determining the total cost associated with execution of the activity is further based on a cost associated with execution of the activity by a second thread.

19. A system comprising:

one or more hardware processors;

one or more non-transitory computer-readable media comprising instructions which, when executed by the one or more hardware processors, causes;

identifying a first thread executing an activity;

determining a set of one or more mutual-exclusion locks held by the first thread while the activity is being executed;

determining a total number of threads waiting for at least one of the set of one or more mutual-exclusion locks held by the first thread during execution of the activity;

stochastically interrupting the first thread executing the activity to determine a set of one or more values representing costs associated with executing the activity;

generating a profile for the activity based at least in part on the set of one or more values representing costs associated with executing the activity;

determining a total cost associated with execution of the activity as a function of at least the set of one or more values representing at least one cost associated with the first thread executing the activity and the total number of threads waiting for at least one of the set of one or more mutual-exclusion locks held by the first thread during execution of the activity; and

implementing, based at least in part on the total cost associated with execution of the activity, a performance optimization by modifying code for executing the activity to reduce mutex wait times caused by execution of the activity.