DYNAMIC ANALYSIS OF PROGRAM EXECUTION TO DISCOVER USAGE CLASSES

Abstract

Dynamically predicting the behavior of applications has the potential to be useful in a variety of application management scenarios, such a job scheduling and detecting potential failure conditions. This research explores the use of machine learning techniques to predict an application’s usage class based on the analysis of its assembly-level instruction trace, working under the premise that processes which are similar in usage class will share similar low-level behavior and functionality. A small catalog of usage classes was developed. A machine learning algorithm was employed to model usage classes and then predict class label for each previously unseen application instruction traces. Various levels of performance were observed depending on the type of trace information and the machine learning algorithm employed.