Autonomous resource management system with recorded evaluations of system performance with scheduler control including knowledge learning function
First Claim
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.
0 Assignments
0 Petitions
Accused Products
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.
167 Citations
28 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. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15)
-
-
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. - View Dependent Claims (17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28)
-
Specification