Accurately predicting system behavior of a managed system using genetic programming

• US 6,453,265 B1
• Filed: 12/28/1999
• Issued: 09/17/2002
• Est. Priority Date: 12/28/1999
• Status: Expired due to Fees

• Current Assignee: Hewlett-Packard Development Company, L.P. (HP Inc.)
• Original Assignee: Hewlett-Packard Company (HP Inc.)
• Inventors: Kirshenbaum, Evan R., Dekhil, Mohamed, Holland, Jerremy
• Primary Examiner(s): Hilten, John S.
• Assistant Examiner(s): Washburn, Douglas N
• Application Number:
  US09/473,793
• Time in Patent Office: 994 Days
• Field of Search: 706/13, 706/21, 187/130, 395/22, 395/23, 395/13, 364/554, 714/26
• US Class Current: 702/181
• CPC Class Codes: G06N 3/126 Evolutionary algorithms, e.g. genetic algorithms or genetic programming

Abstract

A system for monitoring system behavior of a managed/instrumented system is described. The system includes a measurement module coupled to the managed system to generate measurement data of the managed system. The measurement data include past measurement data and current measurement data. A store is also provided that stores the past measurement data. A genetic programming module is coupled to the store to evolve the evolved predictor based on the past measurement data and predetermined effectiveness criteria such that the evolved predictor predicts the system behavior of the managed system when supplied with the current measurement data. The genetic programming module genetically manipulates a set of candidate predictors to produce the evolved predictor. A system for generating the evolved predictor is also described.

Claims

1. A system for monitoring and managing system behavior of a distributed application system, comprising:

a measurement module coupled to the distributed application system to generate measurement data of the distributed application system, wherein the measurement data include past measurement data and current measurement data, wherein the past measurement data include an identified event of interest;

a store that stores the past measurement data of the distributed application system generated by a monitoring module;

a genetic programming module coupled to the store to evolve an evolved predictor based on the past measurement data of the distributed application system and predetermined effectiveness criteria such that the evolved predictor predicts system behavior of the distributed application system when supplied with the current measurement data, wherein the genetic programming module genetically manipulates a set of candidate predictors to produce the evolved predictor;

a control module coupled to receive predictions produced from the evolved predictor when supplied with the current measurement data to control managed distributed application system to adjust itself in accordance with the predictions.

2. The system of claim 1, further comprising a display coupled to the evolving predictor to display predictions produced by the evolving predictor.

3. The system of claim 1, wherein the past and current measurement data include plurality sets of measurement metrics, wherein the evolved predictor is a genetically generated program that can process the current measurement data to predict the behavior of the distributed application system ahead of time.

4. The system of claim 1, wherein each of the set of candidate predictors is described by a plurality of mathematical functions, conditions, constants, and other control operators such that each program represents an algorithm with the measured metrics as input.

5. The system of claim 1, wherein the effectiveness criteria are selected from a group which includes maximizing prediction accuracy, minimizing processing window size, maximizing look-ahead time, and minimizing prediction period in which a problem may occur with certain probability.

6. The system of claim 1, wherein the genetic programming module evolves the evolved predictor by

(I) evaluating each of the candidate predictors in accordance with the effectiveness criteria in order to assign an effectiveness value to each of the candidate predictors;

(II) determining which one of the candidate predictors has the best effectiveness value and if that candidate predictor is good enough to serve as the evolved predictor;

(III) if none of the candidate predictors has the best effectiveness value or if the best effectiveness value is not good enough to serve as the evolved predictor, generating next generation candidate predictors by replacing a randomly selected sub-program of one of the candidate predictors with a sub-program of another program and repeating the steps (I)-(II), wherein the candidate predictors are selected based on their effectiveness values.

7. The system of claim 1, wherein each of the set of candidate predictors is described by a plurality of mathematical functions such that each program represents a mathematical expression over the measured metrics.

8. A system for defining or generating an evolved predictor for predicting system behavior of a distributed application system, comprising:

a measurement module coupled to the distributed application system to generate measurement data of the distributed application system, wherein the measurement data include past measurement data and current measurement data, wherein the past measurement data include an identified event of interest;

a store that stores the past measurement data of the distributed application system;

a genetic programming module coupled to the store to evolve the evolved predictor based on the past measurement data and predetermined effectiveness criteria, wherein the genetic programming module genetically manipulates a set of candidate predictors to produce the evolved predictor that predicts the behavior or future events of the distributed application system when supplied with current measurement data, wherein the evolved predictor does not detect any defect within the distributed application system.

9. The system of claim 8, wherein the past and current measurement data include a plurality sets of measurement metrics, wherein the evolved predictor is a genetically generated program that can process the current measurement data to predict the behavior of the system ahead of time.

10. The system of claim 8, wherein each of the set of candidate predictors is described by a plurality of mathematical functions such that each candidate predictor represents a mathematical expression over the measured metrics.

11. The system of claim 8, wherein each of the set of candidate predictors is described by a plurality of mathematical functions, conditions, constants, and other control operators such that each candidate predictor represents an algorithm with the measured metrics as input.

12. The system of claim 8, wherein the effectiveness criteria are selected from a group which include maximizing prediction accuracy, minimizing processing window size, maximizing the look-ahead time, minimizing the prediction period in which a problem may occur with certain probability.

13. The system of claim 8, wherein the genetic programming module evolves the evolved predictor by

(I) evaluating each of the candidate predictors in accordance with the effectiveness criteria in order to assign an effectiveness value to each of the candidate predictors;

(II) determining which one of the candidate predictors has the best effectiveness value and if that candidate predictor is good enough to serve as the evolved predictor;

(III) if not, generating next generation candidate predictors by replacing a randomly selected sub-program of one of the candidate predictors with a sub-program of another program and repeating the steps (I)-(II), wherein the candidate predictors are selected based on their effectiveness values.