Towards an automated framework for task scheduling

Abstract

the ongoing sh.-inkage of semiconductor geometr'ies allows for increasingly higher system-on-chip (SoC) densities, with more and more on-chip pl·ocessol·s. As a result, task scheduling has become an important concern in system design and reseal'ch in this area has pmduced substantial and dive.-sified knowledge. This paper addl'esses the issue of how to effectively repl'esent and use this knowledge in the context of design automation tools. A new methodology based on functional concept analysis is pl'esented that structul'es the available task scheduling knowledge for optimal application to multiprocessOi' SoC design.

Association rules learning technique for knowledge mining about scheduling algorithm performance

“…In [15], we proposed an automated framework that helps the designer select a scheduling algorithm from a set of alternatives. It relies on creating a repository of uniformly formatted, validated knowledge available from different sources.…”

“…Create a database of applications composed of randomly produced DAGs [9] with attribute definitions as proposed in [15] and relevant performance metrics that are computed after applying the scheduling algorithm to each DAG. The resulting information (DAG attributes, plus metrics) for each application is then stored as a numerical record.…”