Detection of Performance Bottlenecks: A Case-Based Reasoning Approach

摘要

Automating computer performance management has been traditionally treated by rule-based systems, which are less modular than case-based systems in that adding (or removing) a new case in a case-based system does not affect any existing case. It is noted that non-expert users of performance-modelling software, both in preparing inputs and interpreting outputs, tend to use old cases by associating computer systems that reveal similar performance characteristics. A computer system can be described by a set of (attribute, value) pairs. These pairs, which represent classification criteria, enable the users to select a system (a case) from already known systems (case base) based on the degree of similarities between the performance description of a new system and of the selected system. These systems (cases) contain quantitative and qualitative attributes and, therefore, require a special treatment for indexing and retrieval of similar cases. In this work, we present an integrated approach that uses fuzzy set concepts. The approach converts the quantitative attributes into qualitative terms for indexing and retrieval.