Apparatus and method for managing a distributed data processing system workload according to a plurality of distinct processing goal types

• US 5,675,739 A
• Filed: 02/03/1995
• Issued: 10/07/1997
• Est. Priority Date: 02/03/1995
• Status: Expired due to Fees

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Yocom, Peter Bergersen, Eilert, Catherine Krueger
• Primary Examiner(s): Kriess, Kevin A.
• Assistant Examiner(s): Richey, Michael T.
• Application Number:
  US08/383,168
• Time in Patent Office: 977 Days
• Field of Search: 395/650, 395/700, 395/670, 395/671, 395/672, 395/675, 395/180.01, 395/200.03, 395/200.11
• US Class Current: 709/226
• CPC Class Codes: G06F 9/5083 Techniques for rebalancing the load in a distributed system

Abstract

An apparatus for managing a workload distributed across data processing systems in accordance with a common performance standard, which includes a means for measuring the performance of the work units to create local performance data; a means for sending said local performance data to at least one other system; a means for receiving performance data from at least one other system to create remote performance data; and a means responsive to said local and remote performance data for adjusting at least one of the system control parameters to modify the performance of the work units on the system to achieve the common performance standard, is disclosed. Also disclosed is a method for managing workload as represented by the apparatus.

Claims

1. Apparatus for managing a workload comprising work units distributed across a plurality of data processing systems in accordance with a common performance standard, said common performance standard comprising at least one user specified performance goal, each of said systems having access to a stored representation of said performance standard and performing assigned work units in accordance with one or more system control parameters, said apparatus comprising, on each of said systems:

a) means for measuring performance goal achievement of the work units on said system to create local performance goal achievement data;

b) means for sending said local performance goal achievement data to at least one other system in said plurality of systems;

c) means for receiving performance goal achievement data from at least one other system in said plurality of systems to create remote performance goal achievement data; and

d) means responsive to said local and remote performance goal achievement data for adjusting at least one of said system control parameters to modify the performance of said work units on said system to achieve said common performance standard.

2. The apparatus of claim 1 in which said performance standard comprises a common set of user specified performance goals of distinct types, said work units being organized into classes, each of which has a performance goal of one of said types.

3. A method for managing a workload comprising work units distributed across a plurality of data processing systems in accordance with a common performance standard, said common performance standard comprising at least one user specified performance goal, each of said systems having access to a stored representation of said performance standard and performing assigned work units in accordance with one or more system control parameters, said method comprising the steps, performed on each of said systems, of:

a) measuring performance goal achievement of the work units on said system to create local performance goal achievement data;

b) sending said local performance goal achievement data to at least one other system in said plurality of systems;

c) receiving performance goal achievement data from at least one other system in said plurality of systems to create remote performance goal achievement data; and

d) adjusting at least one of said system control parameters in response to said local and remote performance goal achievement data to modify the performance of said work units on said system to achieve said common performance standard.

4. The method of claim 3 in which said performance standard comprises a common set of user specified performance goals of distinct types, said work units being organized into classes, each of which has a performance goal of one of said types.

5. An apparatus for managing a distributed data processing system workload having a common set of user specified performance goals across a plurality of independent and cooperating data processing systems according to one or more distinct performance goal types, said data processing system workload comprising work units organized into one or more performance goal classes, each of said classes having a performance goal from said common set of performance goals falling into one of said performance goal types, each of said data processing systems having access to a stored representation of said user specified performance goals and said apparatus comprising, on each of said data processing systems,

a) workload manager means for accessing said common set of user specified performance goals and creating system control data within a class table means for defining said performance goal classes;

and

b) system resource manager means for managing system resources according to said common set of performance goals, said system resource manager means comprising;

i) sampler means for sampling states of said work units and creating sample data representative of local performance goal achievement;

ii) send data means for sending local performance goal achievement data to at least one other system in said plurality of data processing systems;

iii) remote data receiver means for receiving performance goal achievement data from at least one other system in said plurality of data processing systems; and

iv) multi-system goal driven performance controller means responsive to said sample data, said performance goal achievement data from said at least one other system and said system control data for adjusting at least one system control parameter to cause a change to the performance of said work units to achieve said common set of performance goals.

6. The apparatus of claim 5 in which said class table means further comprises at least one importance parameter assigning a relative importance of a performance goal from said common set of performance goals.

7. The apparatus of claim 5 in which said one or more goal types comprise response time.

8. The apparatus of claim 5 in which said one or more goal types comprise execution velocity.

9. The apparatus of claim 5 in which said multi-system goal driven performance controller means comprises:

a) multi-system performance index calculation means for calculating a multi-system performance index for each of said goal classes across said plurality of data processing systems;

b) local performance index calculation means for calculating a local performance index for each of said goal classes;

c) select receiver means responsive to said calculated multi-system performance indexes and said calculated local performance indexes for selecting a receiver class to be accorded improved service;

d) find bottleneck means responsive to said sample data for identifying a system resource bottleneck affecting said selected receiver class; and

e) fix means for adjusting said system control parameters, said fix means comprising at least one parameter adjustment means, each of said parameter adjustment means being responsive to said identification of an associated system resource bottleneck by said find bottleneck means and adjusting a particular system control parameter associated with said associated system resource bottleneck.

10. The apparatus of claim 6 in which said multi-system goal driven performance controller means comprises:

a) multi-system performance index calculation means for calculating a multi-system performance index for each of said goal classes across said plurality of data processing systems;

b) local performance index calculation means for calculating a local performance index for each of said goal classes;

c) select receiver means for selecting a receiver class to be accorded improved service, said select receiver means responsive to said calculated multi-system performance indexes, said calculated local performance indexes and said at least one importance parameter;

d) find bottleneck means for identifying a system resource bottleneck affecting said selected receiver class, said find bottleneck means being responsive to said sample data; and

e) fix means for adjusting said system control parameters, said fix means comprising at least one parameter adjustment means, each of said parameter adjustment means being responsive to said identification of an associated system resource bottleneck by said find bottleneck means and adjusting a particular system control parameter associated with said associated system resource bottleneck.

11. The apparatus of claim 9 in which said multi-system goal driven performance controller means further comprises assess net value means, responsive to said fix means, for assessing the net effect on performance of a possible change to said particular system control parameter and indicating said net effect to said fix means.

12. The apparatus of claim 9 in which said multi-system goal driven performance controller means further comprises assess net value means, responsive to said fix means, for assessing the net effect on performance of a possible change to said particular system control parameter by using said multi-system performance index and indicating said net effect to said fix means.

13. The apparatus of claim 9 in which said multi-system goal driven performance controller means further comprises assess net value means, responsive to said fix means, for assessing the net effect on performance of a possible change to said particular system control parameter by using said multi-system performance index and said local system performance index and indicating said net effect to said fix means.

14. The apparatus of claim 9 in which said multi-system goal driven performance controller means further comprises select donor means, responsive to said fix means, for selecting a donor class to be accorded degraded service in favor of said selected receiver class.

15. The apparatus of claim 9 in which said system resource bottleneck is identified by said find bottleneck means to be a CPU bottleneck, and in which said fix means comprises adjust dispatch priority means for adjusting donor class dispatch priority and receiver class dispatch priority in response to said identifying of said CPU bottleneck.

16. The apparatus of claim 9 in which said system resource bottleneck is identified by said find bottleneck means to be a multi-program level delay, and in which said fix means comprises increase multi-program level slot means for increasing an multi-program level slot total in response to said identifying of said multi-program level delay.

17. The apparatus of claim 9 in which said system resource bottleneck is identified by said find bottleneck means to be an auxiliary storage swap delay, and in which said fix means comprises protect means for increasing swap protect time for said receiver class.

18. The apparatus of claim 9 in which said system resource bottleneck is identified by said find bottleneck means to be an auxiliary storage paging delay, and in which said fix means comprises increase target means for increasing a storage target in response to said identifying by said find bottleneck means.

19. The apparatus of claim 9 in which said fix means further comprises project multi-system and local performance indexes means for projecting effects of dispatching priority changes on said multi-system and local performance indexes of said donor class and said receiver class.

20. The apparatus of claim 9 in which said fix means further comprises project multi-system and local performance index means for projecting effects of multi-program level slot changes on said multi-system and local performance indexes of said donor class and said receiver class.

21. The apparatus of claim 9 in which said fix means further comprises project multi-system and local performance index means for projecting effects of swap delay changes on said multi-system and local performance indexes of said donor class and said receiver class.

22. The apparatus of claim 9 in which said fix means further comprises project multi-system and local performance index means for projecting effects of auxiliary storage paging delay changes on said multi-system and local performance indexes of said donor class and said receiver class.

23. The apparatus of claim 9 in which said system resource bottleneck is identified by said find bottleneck means to be a CPU bottleneck, and in which said fix means comprises adjust dispatch priority means for adjusting donor class dispatch priority in response to said identifying of said CPU bottleneck.

24. The apparatus of claim 9 in which said system resource bottleneck is identified by said find bottleneck means to be a CPU bottleneck, and in which said fix means comprises adjust dispatch priority means for adjusting receiver class dispatch priority in response to said identifying of said CPU bottleneck.

25. A method for managing a distributed data processing system workload having a common set of user specified performance goals across a plurality of independent and cooperating data processing systems according to one or more distinct performance goals type, said data processing system workload comprising work units organized into one or more performance goals classes, each of said classes having a performance goal from said common set of user specified performance goals falling into one of said performance goal types, each of said data processing systems comprising a stored representation of said common set of performance goals and said method comprising the steps of:

a) accessing said common set of user specified performance goals by each system and creating system control data; and

b) managing system resources according to said common set of performance goals by;

i) periodically sampling states of said work units and creating sample data representative of local performance goal achievement;

ii) sending performance goal achievement data to at least one system in said plurality of data processing system;

iii) receiving performance goal achievement data from at least one system in said plurality of data processing systems;

iv) adjusting at least one system control parameter in response to said sample data and said system control data to change the performance of said work units so as to achieve said common set of performance goals for said particular ones of said work units.

26. The method of claim 25 in which said system control data comprises at least one importance parameter assigning a relative importance to a performance goal from said common set of performance goals.

27. The method of claim 25 in which said work units are organized into a plurality of performance goal classes, said step of adjusting comprising the steps of:

a) calculating a multi-system performance index for each of said classes across said plurality of data processing systems;

b) calculating a local performance index for each of said classes;

c) selecting a receiver class from said plurality of performance goal classes to be accorded improved service, in response to said calculated multi-system and local performance indexes;

d) in response to said sample data, identifying a system resource bottleneck affecting said selected receiver class; and

e) adjusting one or more of said system control parameters in response to said identifying said system resource bottleneck, said one or more system control parameters being associated with said identified system resource bottleneck.

28. The method of claim 27 in which said work units are organized into a plurality of performance goal classes, said step of adjusting comprising the steps of:

a) calculating a multi-system performance index for each of said classes across said plurality of data processing systems;

b) calculating a local performance index for each of said classes;

c) selecting a receiver class from said plurality of performance goal classes to be accorded improved service, in response to said calculated multi-system performance indexes, said calculated local performance indexes and said at least one importance parameter;

d) in response to said sample data, identifying a system resource bottleneck affecting said selected receiver class; and

e) adjusting one or more of said system control parameters in response to said identifying said system resource bottleneck, said one or more system control parameters being associated with said identified system resource bottleneck.

29. The method of claim 27 in which said step of calculating multi-system performance index comprising the steps of

a) determining whether a goal class has a response time goal;

b) if said class has said response time goal;

i) obtaining multi-system response time data;

ii) obtaining local response time data; and

iii) from said multi-system response time data and said local response time data, determining multi-system performance indexes;

c) determining whether a goal class has an execution velocity goal; and

d) if said goal class has said execution velocity goal;

i) obtaining multi-system velocity data;

ii) obtaining local velocity data; and

iii) from said multi-system velocity data and said local velocity data, determining multi-system performance indexes.

30. The method of claim 29 further comprising the step of selecting a donor class to be accorded degraded service in favor of said receiver class.

31. The method of claim 27 further comprising the step of assessing the net effect on performance of a possible change to said one or more system control parameters prior to said adjusting.

32. The method of claim 27 in which said step of identifying a system resource bottleneck identifies a CPU bottleneck, and in which said at least one system control parameter is a dispatching priority parameter associated with said donor class and said receiver class.

33. The method of claim 27 in which said step of identifying a system resource bottleneck identifies a multi-program level delay, and in which said at least one system control parameter is a multi-program level slot which is added, by said step of adjusting, in response to said step of identifying.

34. The method of claim 27 in which said step of identifying a system resource bottleneck identifies an auxiliary storage swap delay, and in which said at least one or more system control parameter is a swap protect time parameter which is increased for said receiver class in response to said step of identifying.

35. The method of claim 27 in which said step of identifying a system resource bottleneck identifies an auxiliary storage paging delay, and in which said at least one system control parameter is a storage target which is increased for said receiver class in response to said step of identifying.

36. The method of claim 27 in which said step of adjusting comprises the step of projecting effects of dispatching priority changes on said multi-system and local performance indexes of said donor class and said receiver class.

37. The method of claim 27 in which said step of adjusting comprises the step of projecting effects of multi-program level slot changes on said multi-system and local performance indexes of said donor class and said receiver class.

38. The method of claim 27 in which said step of adjusting comprises the step of projecting effects of swap delay changes on said multi-system and local performance indexes of said donor class and said receiver class.

39. The method of claim 27 in which said step of adjusting comprises the step of projecting effects of auxiliary storage paging delay changes on said multi-system and local performance indexes of said donor class and said receiver class.

40. The method of claim 27 in which said step of identifying a system resource bottleneck identifies a CPU bottleneck, and in which said at least one system control parameter is a dispatching priority parameter associated with said donor class.

41. The method of claim 27 in which said step of identifying a system resource bottleneck identifies a CPU bottleneck, and in which said at least one system control parameter is a dispatching priority parameter associated with said receiver class.

42. A method for obtaining multi-system performance goal achievement indexes of user specified performance goals for a distributed data processing system workload comprising the steps of:

a) determining if a goal class has a user specified response time goal;

b) if said class has said response time goal;

i) obtaining multi-system response time data;

ii) obtaining local response time data; and

iii) from multi-system response time data and local response time data determining multi-system performance indexes; and

c) determining whether a goal class has a user specified execution velocity goal;

d) if said goal class has said execution velocity goal;

i) obtaining multi-system velocity data;

ii) obtaining local velocity data; and

iii) from said multi-system velocity data and said local velocity data, determining multi-system performance goal achievement indexes.