Automatic Classification of Processes in a General-Purpose Operating System

The scheduler's main goal in a general purpose multitasking operating system is to provide a fair share of processor time to all processes, in order to achieve good performance and an adequate response time for interactive applications. Each process has its own demands for processing and response time, which can not always easily be informed by the user or inferred by the scheduler itself. This article aims to explore the possibilities of applying data mining techniques to the mass of information held by the system kernel for each process, in order to 1) automatically discover groups of processes with similar behavior and 2) automatically classify new processes in these groups. The automatic classification of processes into groups of similar behavior can significantly assist the task of the process scheduler.