Identifying program phase changes through program working set analysis

• US 7,890,933 B2
• Filed: 02/05/2004
• Issued: 02/15/2011
• Est. Priority Date: 02/05/2003
• Status: Active Grant

• Current Assignee: Wisconsin Alumni Research Foundation (University of Wisconsin System)
• Original Assignee: Wisconsin Alumni Research Foundation (University of Wisconsin System)
• Inventors: Dhodapkar, Ashutosh, Smith, James
• Primary Examiner(s): Yigdall; Michael J
• Assistant Examiner(s): Wang; Ben C
• Application Number:
  US10/772,555
• Publication Number:
  US 20040158818A1
• Time in Patent Office: 2,567 Days
• Field of Search: None
• US Class Current: 717/127
• CPC Class Codes: G06F 11/3612 by runtime analysis performance monitoring G06F11/3466

Abstract

The disclosure presents systems and methods to identify program workings sets, detect working set changes and estimate working set sizes. The system generates a highly compressed representation of the working set, called a working set signature, by hashing working set elements into a data structure and setting the entries touched. The working set signature identifies, or is a representation of, the working set. The system can detect a working set change by comparing the signatures of consecutive working sets using a metric called a relative signature distance. The working set size is estimated by counting the number of bits set in the signature. The system can be used to compactly represent various types of working sets such as instruction, data and branch working sets. The system can detect program working set changes (or phase changes) independent of any micro-architectural specification.

Claims

1. An apparatus comprising:

a processing unit of a processor;

a memory coupled to the processor; and

an instruction set operable on the processing unit of the processor and including instructions;

to instantiate a data structure in the memory to collect a representation of a working set; and

to define a hash unit operable on the processing unit to map a plurality of working set elements into the data structure using a hash function,

wherein, in a program, the working set W(t_i, τ

) for i=1, 2 . . . , where i is an integer, is a set of distinct memory segments {s₁, s₂ . . . s_ω

} accessed over the i^th window of size τ

within a time interval t_i;

wherein the window is a sequence of τ

consecutive memory accesses;

wherein the working set size is ω

, the cardinality of the set of unique segments that are accessed by members of the window.

2. The apparatus of claim 1 wherein the data structure is a 2ⁿ×

m bit table, where n is a number of bit table entries and m is a width of the bit table.

3. The apparatus of claim 2 wherein m is in the range of 1 to 64.

4. The apparatus of claim 2 wherein m=1.

5. The apparatus of claim 2 wherein n is in the range of 1 to 20.

6. The apparatus of claim 1 wherein the data structure is a 2ⁿ-bit vector.

7. The apparatus of claim 6 wherein n=1.

8. A computerized method of creating a representation of a working set, the computerized method comprising:

mapping a plurality of working set elements into fields of a data structure using a hash function,

wherein, in a program, the working set W(t_i, τ

) for i=1, 2 . . . , where i is an integer, is a set of distinct memory segments {s₁, s₂ . . . s_ω

} accessed over the i^th window of size τ

within a time interval t_i;

wherein the window is a sequence of τ

consecutive memory accesses;

wherein the working set size is ω

, the cardinality of the set of unique segments that are accessed by members of the window.

9. The computerized method of claim 8 wherein the mapping is performed for a fixed interval of program execution.

10. The computerized method of claim 9 wherein the data structure is reset prior to each fixed interval of program execution.

11. The computerized method of claim 10 further comprising saving the fields of the data structure prior to resetting the data structure.

12. A computerized method of creating a representation of a working set, the computerized method comprising:

executing a program for a fixed interval, the program comprising instructions identified by a program counter;

performing a hash function on the program counter to create a hash value for each instruction executed during the fixed interval; and

updating a field of a table indexed by the hash value wherein the table represents the working set,

wherein, in a second program, the working set W(t_i, τ

) for i=1, 2 . . . , where i is an integer, is a set of distinct memory segments {s₁, s₂ . . . s_ω

} accessed over the i^th window of size τ

within a time interval t_i;

wherein the window is a sequence of τ

consecutive memory accesses;

wherein the working set size is ω

, the cardinality of the set of unique segments that are accessed by members of the window.

13. A computer system comprising:

a bus;

a memory coupled to the bus; and

a processor coupled to the memory and the bus;

the processor comprising;

a data structure to collect a representation of a working set; and

a hash unit to map a plurality of working set elements into the data structure using a hash function,

wherein, in a program, the working set W(t_i, τ

) for i=1, 2 . . . , where i is an integer, is a set of distinct memory segments {s₁, s₂ . . . s_ω

} accessed over the i^th window of size τ

within a time interval t_i;

wherein the window is a sequence of τ

consecutive memory accesses;

wherein the working set size is ω

, the cardinality of the set of unique segments that are accessed by members of the window.

14. The computer system of claim 13, further comprising:

an instruction retirement unit; and

wherein the data structure and the hash unit are part of an instruction retirement unit.

15. A computerized method of estimating size of a working set, the method comprising:

receiving a signature for a working set; and

estimating the size of the working set based on the size of the signature,

wherein, in a program, the working set W(t_i, τ

) for i=1, 2 . . . , where i is an integer, is a set of distinct memory segments {s₁, s₂ . . . s_ω

} accessed over the i^th window of size τ

within a time interval t_i;

wherein the window is a sequence of τ

consecutive memory accesses;

wherein the working set size is ω

, the cardinality of the set of unique segments that are accessed by members of the window.

16. The computerized method of claim 15 wherein the estimating is performed with the following function:

17. A computerized method of detecting working set changes, the method comprising:

comparing a current working set signature to a previous working set signature;

calculating a relative signature distance between the current working set signature and the previous working set signature; and

identify a working set change when the relative signature distance exceeds a predetermined threshold,

wherein, in a program, the working set W(t_i, τ

) for i=1, 2 . . . , where i is an integer, is a set of distinct memory segments {s₁, s₂ . . . s_ω

} accessed over the i^th window of size τ

within a time interval t_i;

wherein the window is a sequence of τ

consecutive memory accesses;

wherein the working set size is ω

, the cardinality of the set of unique segments that are accessed by members of the window.

18. The computerized method of claim 17 wherein the working set change indicates a phase change in a program.

19. A computerized method of identifying a recurring working set, the method comprising:

comparing a current working set signature to one or more previous working set signatures;

calculating a relative signature distance between the current working set signature and the one or more previous working set signatures; and

identifying a recurring working set when the relative signature distance between the current working set signature and one of the previous working set signatures is within a predetermined threshold,

wherein, in a program, the working set W(t_i, τ

) for i=1, 2 . . . , where i is an integer, is a set of distinct memory segments {s₁, s₂ . . . s_ω

} accessed over the i^th window of size τ

within a time interval t_i;

wherein the window is a sequence of τ

consecutive memory accesses;

wherein the working set size is ω

, the cardinality of the set of unique segments that are accessed by members of the window.

20. The computerized method of claim 19 further comprising identifying a new working set when the relative signature distance between the current working set signature the one or more previous working set signatures exceeds a predetermined threshold.

21. The computerized method of claim 20 further comprising maintaining a table of the one or more previous working set signatures.

22. A hardware reconfiguration method comprising:

maintaining a table comprising a plurality of working set signatures for a program;

upon detecting a working set change, looking up a working set signature for a current working set in the table;

if the working set signature is in the table, reinstating a hardware configuration for the current working set; and

if the working set signature is not in the table;

identifying a new hardware configuration for the current working set and saving the working set signature and the new hardware configuration,

wherein, in a program, the working set W(t_i, τ

) for i=1, 2 . . . , where i is an integer, is a set of distinct memory segments {s₁, s₂ . . . s_ω

} accessed over the i^th window of size τ

within a time interval t_i;

wherein the window is a sequence of τ

consecutive memory accesses;

wherein the working set size is ω

, the cardinality of the set of unique segments that are accessed by members of the window.

23. The method of claim 22 wherein the working set change indicates a phase change in a program.