Automating performance management using case-based reasoning

Abstract

Resolving performance problems in information systems is a time consuming task for which automation can be of great benefit. Traditionally, this automation has been provided by rule-based systems. Herein is described an alternative approach that employs case-based reasoning (CBR). CBR matches new problem instances against previously solved problems (or cases), retrieving only those cases that are most similar to the problem instance. Case-based reasoning has the disadvantage of being less flexible than rules due to several hard-to-change decisions that are made at design time: the choice of attributes to match, the encodings used for attribute values, and the similarity metric employed. Also, compared with rules, CBR is less adept at explaining its conclusions. However, CBR is much more modular than rule-based approaches in that adding (or removing) a new case does not affect any existing case; in contrast, modifying a rule-base is, in general, a complex task. The modularity of CBR systems enables performance management automation that is custom tailored by end-users based on their own experience. Such automation can empower end-users to solve their own performance problems, thereby freeing scarce support staff for other tasks. However, realizing these benefits in practice requires considerations for user-interfaces, software architecture, and business processes.