Performance improvement tool for rule based expert systems

• US 5,129,043 A
• Filed: 08/14/1989
• Issued: 07/07/1992
• Est. Priority Date: 08/14/1989
• Status: Expired due to Fees

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Yue, Po C.
• Primary Examiner(s): MacDonald, Allen R.
• Assistant Examiner(s): Davis, George
• Application Number:
  US07/393,526
• Time in Patent Office: 1,058 Days
• Field of Search: 395/51, 364/478, 364/200, 364/471, 364/184, 364/507
• US Class Current: 706/60
• CPC Class Codes: G06F 11/3466 Performance evaluation by tracing or monitoring
• CPC Class Codes: G06N 5/022 Knowledge engineering; Knowledge acquisition

Abstract

A monitoring system monitors an expert system application while such application executes for a period of time. The monitoring system records selected performance data relative to the operating efficiency of the expert system, and writes such data into a database for later use. After a test run of the expert system is completed, the monitoring system can select a small number of candidate rules for revision by a user, and display data to the user which allows the user to determine what types of changes need to be made to the candidate rules and to the working memory data organization.

Claims

1. A system for monitoring the performance of a rule-base application having rules and a working memory containing elements, each rule therein having a left-hand-side and a right-hand-side, comprising:

a timer for accumulating match time caused by executing rule right-hand-sides;

a first counter for counting working memory operations performed for each class of elements in the working memory;

a second counter for counting the number of times each condition test is performed, and separately counting the number of times each condition test is successful;

a third counter for counting the number of working memory element combinations which satisfy a pattern each time that pattern is matched; and

means for storing information generated by said counters and said timer into a database.

2. The system of claim 1, further comprising a second timer for accumulating execution time of the rule right-hand-sides.

3. The system of claim 1, further comprising means for generating reports indicative of the information stored in the database.

4. The system of claim 1, further comprising:

means for selecting suspect rules for examination; and

means for displaying information extracted from the database concerning the suspect rules.

5. The system of claim 4, wherein said selecting means comprises:

a comparator for determining a culprit rule whose right-hand-side, upon execution, caused a relatively large amount of time to be spent in a matching cycles;

a first selector for selecting working memory classes affected by execution of the culprit rule right-hand-side; and

a second selector for selecting all rules having left-hand-sides affected by changes in the selected working memory classes, such selected rules defining the suspect rules.

6. A method for providing information on causes of performance inefficiences of a rule-based expert system, comprising the steps of:

accessing performance data about the expert system generated during operation of the expert system;

processing the accessed performance data to determine a culprit rule which caused a relatively large amount of time to be consumed in matching cycles as a result of executing its right-hand-side;

selecting working memory classes affected by execution of the culprit right-hand-side; and

selecting as suspect rules all rules having left-hand-sides affected by changes in the selected working memory classes; and

displaying performance data for the suspect rules.

8. The method of claim 6, further comprising the step of accumulating execution times for rule right-hand-sides.

9. The method of claim 6, further comprising the steps of:

accessing the database;

selecting a set of suspect rules; and

displaying performance data for the suspect rules.

10. The method of claim 8, wherein said selecting step comprises the steps of:

processing the accessed performance data to determine a culprit rule which caused a relatively large amount of time to be consumed in matching cycles as a result of executing its right-hand-side;

selecting working memory classes affected by execution of the culprit right-hand-side; and

selecting as suspect rules all rules having left-hand-sides affected by changes in the selected working memory classes.

7. A method for monitoring the performance of a rule-based application, comprising the steps of:

accumulating match times caused by executing rule right-hand-sides;

counting working memory operations performed for each class of elements in a working memory;

counting how often each condition test in a rule left-hand-side is performed;

counting how often each condition test is successfully performed;

counting how many working memory elements satisfy a pattern in a rule left-hand-side when such pattern is matched; and

storing information generated by such accumulating and counting steps into a database.