Method of controlling the performance of a data storage system

• US 7,281,097 B1
• Filed: 06/30/2004
• Issued: 10/09/2007
• Est. Priority Date: 06/30/2004
• Status: Active Grant

• Current Assignee: Emc IP Holding Company LLC (Dell Technologies Inc.)
• Original Assignee: EMC Corporation (Dell Technologies Inc.)
• Inventors: Norgren, Matthew, Madhukar, Manish, Lawson, Malcolm, Matulioniene, Rasa
• Primary Examiner(s): Bragdon; Reginald
• Assistant Examiner(s): Flournoy; Horace L.
• Application Number:
  US10/881,530
• Time in Patent Office: 1,196 Days
• Field of Search: None
• US Class Current: 711/154
• CPC Class Codes: G06F 11/3457 Performance evaluation by simulation
• CPC Class Codes: G06F 3/0613 in relation to throughput
• CPC Class Codes: G06F 3/0632 by initialisation or re-initialisation of storage systems
• CPC Class Codes: G06F 3/0656 Data buffering arrangements
• CPC Class Codes: G06F 3/0689 Disk arrays, e.g. RAID, JBOD

Abstract

A method for optimizing the operation of a data storage system utilizes a genetic algorithm to adjust internal parameters of the system. The method involves determining a set of optimum values of each of two or more different data array parameters, such as data request types that will be processed concurrently, to achieve performance goals that are set by the user for the system. The optimum values are determined by iteratively evaluating the performance of the system while it processes the different combinations of values of the two or more data requests, saving the combinations of numbers that yield performance parameters that approach the goal and eliminating combinations of number of data request types that do not.

Claims

1. A method of controlling the performance of a data storage system which concurrently processes a number N of data requests, of which, a number A are of a first data request type and a number B are of a second data request type, the method comprising:

(A) setting a performance goal for the data storage system; and

(B) determining, through the use of a genetic algorithm, an optimum number A_O of the first data request type and an optimum number B_O of the second data request type to concurrently process, to achieve the performance goal for the data storage system; and

operating the data storage system to concurrently process the number A_O first data request types and the number B_O second data request types to cause the data storage system to perform in accordance with the performance goal.

2. The method of claim 1 wherein Step B comprises:

(C) randomly selecting a set of solutions, each solution including initial numbers A_I of the first data request type and B_I of the second data request type;

(D) evaluating each solution in the set for a period of time by processing the number A_I of first data request types and B_I second data request types, measuring the performance of the data storage system, comparing the measured performance to the performance goal and assigning a fitness value to the solution based on the comparison, wherein solutions having a greater fitness value yield a performance closer to the performance goal than those having a lesser fitness value;

(E) selecting two of the solutions from the set by performing a weighted random selection, based on the fitness value assigned to each solution;

(F) combining the two selected solutions to obtain a new solution including a number A_N of first data request types and a number B_N of second data request types;

(G) evaluating the new solution by processing the number A_N of first data request types and B_N second data request types, measuring the performance parameter, comparing the measured performance to the performance goal and assigning a fitness value to the new solution based on the comparison;

(H) including the new solution in the set of solutions;

(I) eliminating the solution in the set having the lowest fitness value; and

(J) repeating steps (E) through (I) to generate a set of solutions having increased fitness values, until the optimum number A_O of the first data request type and the optimum number B_O of the second data request type are determined.

3. The method of claim 2 further comprising, in less than all of the new solutions, randomly altering the new solution after step (F).

4. The method of claim 3 wherein the randomly altering step comprises applying a mutation function to the new solution.

5. The method of claim 4, wherein the random altering step is performed in less than approximately 10% of the new solutions.

6. The method of claim 5, wherein the random altering step is performed in approximately 5% of the new solutions.

7. The method of claim 2 wherein the set of solutions includes a first list of solutions for the first data request type and a second list of solutions for the second data request type.

8. The method of claim 7 wherein, in Step E, one solution is selected from each of the first and second lists in the set.

9. The method of claim 7 wherein fitness values assigned to solutions in the first list are weighed differently than fitness values assigned to solutions in the second list.

10. The method of claim 1 wherein the performance parameter includes one of inputs/outputs per second, a response time for a request type and a throughput amount for a request type.

11. The method of claim 1 wherein A+B=N.

12. A method of controlling the performance of a data storage system which concurrently processes a number N of data requests, of which, a number A are of a first data request type and a number B are of a second data request type, the method comprising:

(A) setting a performance goal for the data storage system; and

(B) randomly selecting a set of solutions, each solution including initial numbers A_I of the first data request type and B_I of the second data request type;

(C) evaluating each solution in the set for a period of time by processing the number A_I of first data request types and B_I second data request types, measuring the performance of the data storage system, comparing the measured performance to the performance goal and assigning a fitness value to the solution based on the comparison, wherein solutions having a greater fitness value yield a performance closer to the performance goal than those having a lesser fitness value;

(D) selecting two of the solutions from the set by performing a weighted random selection, based on the fitness value assigned to each solution;

(E) combining the two selected solutions to obtain a new solution including a number A_N of first data request types and number B_N of second data request types;

(F) evaluating the new solution by processing the number A_N of first data request types and B_N second data request types, measuring the performance of the data storage system, comparing the measured performance to the performance goal and assigning a fitness value to the new solution based on the comparison;

(G) including the new solution in the set of solutions;

(H) eliminating the solution in the set having the lowest fitness value;

(I) repeating steps (D) through (H) to generate a set of solutions having increased fitness values, until the optimum number A_O of the first data request type and the optimum number B_O of the second data request type are determined; and

(J) operating the data storage system to concurrently process the number A_O first data request types and the number B_O second data request types to cause the data storage system to perform in accordance with the performance goal.

13. The method of claim 12 further comprising, in less than all of the new solutions, randomly altering the new solution after step (F).

14. The method of claim 13 wherein the randomly altering step comprises applying a mutation function to the new solution.

15. The method of claim 13, wherein the random altering step is performed in less than approximately 10% of the new solutions.

16. The method of claim 15, wherein the random altering step is performed in approximately 5% of the new solutions.

17. The method of claim 12 wherein the performance parameter includes one of inputs/outputs per second, a response time for a request type and a throughput amount for a request type.

18. The method of claim 12 wherein A+B=N.

19. The method of claim 12 wherein the set of solutions includes a first list of solutions for the first data request type and a second list of solutions for the second data request type.

20. The method of claim 19 wherein, in Step D, one solution is selected from each of the first and second lists in the set.

21. The method of claim 19 wherein fitness values assigned to solutions in the first list are weighted differently than fitness values assigned to solutions in the second list.

22. A method of controlling the performance of a data storage system which concurrently processes a number N of data requests, of which, a number A are of a first data request type, a number B are of a second data request type and a number C are of data request types other than the first and second data request types, the method comprising:

(A) setting a performance goal for the data storage system;

(B) randomly selecting a set of solutions, each solution including initial numbers A_I of the first data request type, B_I of the second data request type and C_I of the other data request types;

(C) evaluating each solution in the set for a period of time by processing the number A_I of first data request types, B_I second data request types and C_I of the other data request types, measuring the performance of the data storage system, comparing the measured performance to the goal and assigning a fitness value to the solution based on the comparison, wherein solutions having a greater fitness value yield a performance closer to the performance goal than those having a lesser fitness value;

(D) selecting two of the solutions from the set by performing a weighted random selection, based on the fitness value assigned to each solution;

(E) combining the two selected solutions to obtain a new solution including a number A_N of first data request types, a number B_N of second data request types and a number C_N of other data request types;

(F) evaluating the new solution by processing the number A_N of first data request types, the number B_N of second data request types and the number C_N of other data request types, measuring the performance of the data storage system, comparing the measured performance to the performance goal and assigning a fitness value to the new solution based on the comparison;

(G) including the new solution in the set of solutions;

(H) eliminating the solution in the set having the lowest fitness value; and

(I) repeating steps (D) through (H) to generate a set of solutions having increased fitness values, until an optimum number A_O of the first data request type, an optimum number B_O of the second data request type and an optimum number C_O of the other data request types are determined; and

(J) operating the data storage system to concurrently process the number A_O first data request types, the number B_O second data request types and the number C_O other data request types to cause the data storage system to perform in accordance with the performance goal.

23. The method of claim 22 further comprising, in less than all of the new solutions, randomly altering the new solution after step (F).

24. The method of claim 23 wherein the randomly altering step comprises applying a mutation function to the new solution.

25. The method of claim 23, wherein the random altering step is performed in less than approximately 10% of the new solutions.

26. The method of claim 25, wherein the random altering step is performed in approximately 5% of the new solutions.

27. The method of claim 22 wherein the performance parameter includes one of inputs/outputs per second, a response time for a request type and a throughput amount for a request type.

28. The method of claim 22 wherein A+B+C=N.

29. The method of claim 22 wherein the set of solutions includes a first list of solutions for the first data request type, a second list of solutions for the second data request type and a third list of solutions for the other data request types.

30. The method of claim 29 wherein, in Step D, one solution is selected from two of the first, second and third lists in the set.

31. The method of claim 29 wherein fitness values assigned to solutions in the first list are weighted differently than fitness values assigned to solutions in the second list.

32. A computer program product, embodied on a computer readable medium, having computer readable program code thereon for controlling the performance of a data storage system including a storage processing unit which concurrently processes a number N of data requests, of which, a number A are of a first data request type and a number B are of a second data request type, the computer readable program code determining, through the use of a genetic algorithm, an optimum number A_O of the first data request type and an optimum number B_O of the second data request type for the storage processing unit to concurrently process, to enable the data storage system to operate at a performance goal for the data storage system.

33. The computer program product of claim 32, wherein the computer readable program code:

(A) randomly selects a set of solutions, each solution including initial numbers A_I of the first data request type and B_I of the second data request type;

(B) evaluates each solution in the set for a period of time by processing the number A_I of first data request types and B_I second data request types, measuring the performance of the data storage system, comparing the measured performance to the performance goal and assigning a fitness value to the solution based on the comparison, wherein solutions having a greater fitness value yield a performance closer to the performance goal than those having a lesser fitness value;

(C) selects two of the solutions from the set by performing a weighted random selection, based on the fitness value assigned to each solution;

(D) combines the two selected solutions to obtain a new solution including a number A_N of first data request types and a number B_N of second data request types;

(E) evaluates the new solution by processing the number A_N of first data request types and the number B_N of second data request types, measuring the performance of the data storage system, comparing the measured performance to the performance goal and assigning a fitness value to the new solution based on the comparison;

(F) includes the new solution in the set of solutions;

(G) eliminates the solution in the set having the lowest fitness value; and

(H) repeats steps (C) through (G) to generate a set of solutions having increased fitness values, until the optimum number A_O of the first data request type and the optimum number B_O of the second data request type are determined;

the data storage system being controlled to concurrently process the number A_O first data request types and the number B_O second data request types to cause the data storage system to perform in accordance with the performance goal.

34. The computer program product of claim 33 further comprising, in less than all of the new solutions, randomly altering the new solution after step (D).

35. The computer program product of claim 34 wherein the randomly altering step comprise applying a mutation function to the new solution.

36. The computer program product of claim 34, wherein the random altering step is performed in less than approximately 10% of the new solutions.

37. The computer program product of claim 36, wherein the random altering step is performed in approximately 5% of the new solutions.

38. The computer program product of claim 33 wherein the performance parameter includes one of inputs/outputs per second, a response time for a request type and a throughput amount for a request type.

39. The computer program product of claim 33 wherein A+B=N.

40. The computer program product of claim 33 wherein the set of solutions includes a first list of solutions for the first data request type and a second list of solutions for the second data request type.

41. The computer program product of claim 40 wherein fitness values assigned to solutions in the first list are weighted differently than fitness values assigned to solutions in the second list.