| Summary: | Computer performance has become increasingly important at modern computer installations. At the same time, performance analysts have shown a strong interest in using expert system technology to meet these new demands. Because of the importance of the system sort product in improving overall system performance, a prototype expert system (KBSORT) was developed. When embedded in the system sort utility, KBSORT optimizes the CPU time performance of the sort applications generated by the system utility.
A thorough search of the available literature revealed that no work had yet been done on applying expert system technology to enhancing sort performance. Because the research for this study is a first attempt in this area, a rigorous twelve step procedure was followed in developing the KBSORT prototype. The steps included formulating test cases, researching sort performance enhancement techniques, interviewing experts on sort performance, developing the rules for the expert system, embedding the prototype in the sort utility, and running the sort utility to study the difference in CPU time performance when the expert system was in use.
Interviews with sort performance experts revealed that the expert system had to optimize in three areas. It had to choose (1) the best sorting algorithm for the current circumstances, (2) the optimal amount of core for the particular sort, and (3) the proper sort work allocation on the input/output devices available while taking into account the current level of contention for those devices.
The results of this study support the original hypothesis that the use of a sort utility with an embedded expert system designed to improve CPU time performance generates applications that run more efficiently than sort applications generated by a sort utility without such an embedded expert system.
|
|---|
