Autonomous resource management system with recorded evaluations of system performance with scheduler control including knowledge learning function

• US 4,890,227 A
• Filed: 08/25/1987
• Issued: 12/26/1989
• Est. Priority Date: 07/20/1983
• Status: Expired due to Term

• Current Assignee: Hitachi, Ltd.
• Original Assignee: Hitachi, Ltd.
• Inventors: Motoyama, Hiroshi, Yakata, Teruo, Watanabe, Yoshitaka, Sasaki, Koji, Watanabe, Toshinori, Haruna, Koichi, Nishigaki, Toru
• Primary Examiner(s): Anderson, Lawrence E.
• Application Number:
  US07/089,919
• Time in Patent Office: 854 Days
• Field of Search: 364/200 MS File, 364/900 MS File, 364/300
• US Class Current: 706/14
• CPC Class Codes: G06F 11/34 Recording or statistical evaluation of computer activity, e.g. of down time, of input/output operation ; Recording or statistical evaluation of user activity, e.g. usability assessment
• CPC Class Codes: G06F 11/3433 for load management allocation of a server based on load conditions G06F9/505; load rebalancing G06F9/5083; redistributing the load in a network by a load balancer H04L67/1029
• CPC Class Codes: G06F 9/50 Allocation of resources, e.g. of the central processing unit [CPU]
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: Y10S 706/908 Electronic or computer, internal or network, circuit

Abstract

An autonomous resource management method for improving performance of a system in which plurality of process units compete for resources. Previously determined allocation policies are stored in memory along with system status data, control data and an evalution of the effectiveness of the allocation policy. During the operation of the system, the status is detected and an allocation policy is retrieved from memory based on the detected status. The allocation policy is then modified and sent to the scheduler. The effect of modifying the allocation policy is detected and an evaluation of the modified allocation policy is stored in memory. Inferior allocation policies are deleted from memory.

Claims

1. A resource management method for a computer system having resources and in which a plurality of process units compete for utilization of resources, a scheduler to control the resource utilization, a sensor to detect a system status including resource utilization status and a memory, the method to be performed during normal operation of the computer system without operator intervention comprising the steps of:

storing sets of control knowledge data in said memory, each set of control knowledge data including status data representative of a system status as detected by said sensor, control data for controlling said scheduler determined on the basis of said status data of the set and data representative of an effect of applying said control data of the set to said scheduler;

selecting one set of said control knowledge data whose status data substantially corresponds to current status data detected by said sensor to obtain an improved control data applying effect;

modifying control data contained in said selected set of control knowledge data and sending the modified control data to said scheduler;

detecting an effect of applying said modified control data to said scheduler for determining application effect data representative of an effect of applying said modified control data to said scheduler; and

updating said stored control knowledge data on the basis of said current status data, the modified control data corresponding to said current status data and the application effect data resulting from applying said modified control data to said scheduler, whereby overall system performance for similar system status is improved with time.

2. A method according to claim 1, wherein said control data includes a planned utilization ratio of each respective resource for each of a plurality of categories of said process units.

3. A method according to claim 1, wherein said modification of the control data is made by use of random numbers.

4. A method according to claim 1, wherein said status data is represented by at least one of (a) a number of process units waiting to be processed by said system, (b) a trend of increase or decrease of said number of process units and (c) utilization factors of said resources.

5. A method according to claim 4, wherein said status data is represented by a variable average value of said number of process units, said trend of increase or decrease of said number of process units and said utilization factors of said resources.

6. A method according to claim 1, wherein said application effect data is expressed by a function of an average expansion rate of each process unit and a throughput of said system.

7. A method according to claim 6, wherein said application effect data is represented by a variable average value of detected values of said process unit average expansion rate and said throughput.

8. A method according to claim 6, wherein said application effect data is provided in the form of a lookup table.

9. A method according to claim 1, wherein said selection step includes a selection of one set of control knowledge data based on status data and application effect data of each set of control knowledge data and the detected status data.

10. A method according to claim 9, wherein selection of one set of control knowledge data is made by means of a retrieving function having an input variable which includes the detected current status data and a function value corresponding to said application effect data, said retrieving function being imparted with at least one of a plurality of characteristics including (1) an expected value of the application effect data which decreases as said detected current status data is more greatly spaced from the status data in the selected set of control knowledge data, (2) said expected value decreases as the value of said application effect decreases and (3) said expected value decreases each time said selected set of control knowledge data is updated.

11. A method according to claim 10, wherein said retrieving function is defined by a function:

##EQU4## where Y;

a vector of the current detected status data

Y;

a vector of the status data of the set of control knowledge data

J₁ ;

said application effect data

t;

the number of updating of said set of control knowledge data

r₁ ;

a constant vector

α

;

a constant

J_max ;

a constant

A(r₁, Y-Y);

a distance function between two vectors Y and Y having a weight r₁.

12. A method according to claim 1, wherein said updating includes, if said area in said memory has a vacant space, storing a new set of control knowledge data in said vacant space and, if said area has no vacant space, at least one set of said control knowledge data is replaced by said new set of control knowledge data.

13. A method according to claim 12, wherein said replacing is carried out on a basis of a deletability factor of each set of control knowledge data.

14. A method according to claim 13, wherein said deletability is imparted with at least one of a plurality of characteristics including (1) an expected value of the deletability decreases as said current detected status data is more remote from the status data in the set of control knowledge data, (2) said expected value of the deletability increases as the value of said application effect data decreases and (3) said expected value increases each time said set of control knowledge data is updated.

15. A method according to claim 14, wherein said deletability is defined by a function:

##EQU5## where Y;

a vector of the current detected status data

Y;

a vector of the status data in the set of control knowledge data

J₂ ;

an application effect data

t;

the number of updatings of said set of control knowledge data

r₂ ;

a constant vector

α

;

a constant

J_max ;

a constant

(r₂, Y-Y);

a distance function between two vectors Y and Y having a weight r₂.

16. An autonomous resource management method for controlling the operation of a system in which a plurality of process units compete for resources in the system, said method being performed within said system without operator intervention and comprising the steps of:

storing in memory a knowledge base in the form of electrical codes containing information on a plurality of allocation policies for allocating said process units to said system resources and information as to an evaluation of each performance of said system operated under respective ones of said allocation policies, each allocation policy setting a different policy;

detecting the status of at least one of said process units and said system resources;

retrieving from said knowledge base an allocation policy in the form of an electrical code for executing said process units on the basis of said detected status;

modifying said retrieved allocation policy to generate a modified electrical code;

sending said modified electrical code indicating said modified allocation policy to said system;

allocating said process units to said resources in accordance with said modified electrical code indicating said modified allocation policy;

operating said system to execute said process units allocated to said resources and outputting, from said system, results of said execution in the form of electrical signals;

evaluating a performance of the system operated under the modified allocation policy on the basis of said results; and

updating said knowledge base in memory by writing in said knowledge base said modified electrical codes representing said modified allocation policy and the evaluation of the performance corresponding to said modified allocation policy by deleting the electrical codes representing an inferior allocation policy from said knowledge base in memory.

17. An autonomous resource management method according to claim 16, wherein said electrical codes include information on the status of at least one of said system and said process units.

18. An autonomous resource management method according to claim 16, wherein said system is a multi-programming computer system.

19. An autonomous resource management method according to claim 16, wherein said information in said knowledge base is expressed in the form of electrical codes representing process unit status, computer status, evaluation point data, territory function parameter data and control signal data.

20. An autonomous resource management method according to claim 16, wherein said updating step includes a step for replacing electrical codes representing an inferior allocation policy and an evaluation corresponding to said inferior allocation policy in said knowledge base by electrical codes representing said modified allocation policy and said evaluation corresponding to said modified allocation policy.

21. An autonomous resource management method according to claim 20, wherein said replacing of said inferior allocation policy and the evaluation corresponding to said inferior allocation policy is determined on the basis of another criterion such that an allocation policy having a difference from said another criterion greater than other allocation policies as determined from said detected status is less likely to be substituted.

22. An autonomous resource management method according to claim 21, wherein said criterion is defined by a function, the variables of which include at least ages representative of time periods during which candidate inferior allocation policies being considered have been held in said knowledge base, evaluations of the candidate inferior allocation policy and the gap between said detected status and the status of said candidate inferior allocation policies.

23. An autonomous resource management method according to claim 16, wherein said retrieved allocation policy is determined on the basis of criterion such that an allocation policy having a difference from said criterion greater than other allocation policies as determined from said detected status is more unlikely to be recalled from said knowledge base.

24. An autonomous resource management method according to claim 23, wherein said criterion is defined by a function including variables, the variables of said function including ages representative of time periods during which said electrical codes representing said allocation policies are held in said knowledge base in memory, evaluations of said allocation policies and the age between said detected status and the status of candidate allocation policies.

25. An autonomous resource management method according to claim 23, wherein said criterion includes as parameters thereof ages representative of a time period during which said electrical codes representing said allocation policies are held in said knowledge base in memory so that an older allocation policy is more unlikely to be recalled and more likely to be eliminated from said knowledge base in memory.

26. An autonomous resource management method according to claim 23, wherein the degree of the unlikelihood of an allocation policy in said knowledge base being recalled is modifiable.

27. An autonomous resource management method according to claim 16, wherein said allocation policy to be modified includes electrical codes representing a resource utilization matrix function.

28. An autonomous resource management method according to claim 27, wherein said allocation policy includes electrical codes representing a control time and a parameter for monitoring the status of a computer.