Using static code analysis for improvement of job data availability in bobox task scheduling

Abstract

This paper proposes an optimization method for a task-based parallel framework called Bobox. The method is based on a static code analysis; it reduces a number of short running tasks. Since the Bobox scheduler has only limited runtime information, it can schedule a task even though the task does not have all input data. Such short-running tasks cause an unnecessary scheduling overhead. To achieve more effective scheduling, the proposed optimizer analyzes usage of inputs to remove short running tasks.